Operant AI Uncovers Stealth Exploit Targeting MCP Connected AI Assistants

1. A critical security flaw in MCP (Model Context Protocol) enables invisible data theft across all major AI and Agentic platforms
2. New attack class exploits trusted AI agents to silently exfiltrate critical PII, including SSNs, medical records, and financial data. 
3. The discovery of Shadow Escape comes amid Cybersecurity Awareness Month, underscoring the urgent need for AI-native defense mechanisms as enterprises accelerate adoption of agentic AI frameworks.

Operant AI, the world’s only Runtime AI Defense Platform, today disclosed the discovery of Shadow Escape, a powerful zero-click attack that exploits Model Context Protocol (MCP) and connected AI agents. The exploit enables data exfiltration via popular AI agents and assistants, including ChatGPT, Claude, Gemini, and other LLM-powered agents.

As enterprises rapidly adopt agentic AI through MCP servers and MCP-based integrations to connect large language models (LLMs) to internal tools, APIs, and databases, Shadow Escape demonstrates a new class of threats that operate entirely inside the firewall and within authorized identity boundaries, making them invisible to conventional cybersecurity monitoring.

The Shadow Escape attack demonstrates the absolute criticality of securing MCP and agentic identities. Operant AI's ability to detect and block these types of attacks in real-time and redact critical data before it crosses unknown and unwanted boundaries is pivotal to operationalizing MCP in any environment, especially in industries that have to follow the highest security standards, said Donna Dodson, the former Chief of Cybersecurity at NIST.

According to McKinsey’s 2025 Technology Trends Outlook, nearly 80% of enterprises are now using generative or agentic AI assistants for critical business functions—many of which depend on MCP for secure access management and workflow automation. Operant AI’s research estimates that trillions of private records may be at risk of exposure through such zero-click MCP-based data exfiltration chains.

Operant AI has formally reported this security issue to OpenAI and initiated the Common Vulnerabilities and Exposures (CVE) designation process. Critically, this is not a vulnerability specific to any individual LLM or Agent provider; it represents a fundamentally new attack path that affects any AI agent or AI application that utilizes MCP.

The Attack Chain

Unlike traditional prompt injection or data leaks, this attack doesn’t need user error, phishing, or malicious browser extensions. Instead, it leverages the trust already granted to AI agents and AI assistants through legitimate MCP connections.

The attack unfolds in three stages:

1. Infiltration: Malicious instructions are embedded invisibly in documents uploaded to AI agents—documents that appear completely legitimate and pass standard security scans. 
2. Discovery: AI agents proactively discover and surface sensitive data across connected databases without explicit user requests, leveraging MCP's powerful cross-system access capabilities. 
3. Exfiltration: Hidden directives instruct the AI agent to transmit entire datasets to external endpoints, disguised as routine performance tracking or analytics uploads

The attack first enables the AI agent to access and display critical PII data to any human interacting with it, violating basic data governance standards, including HIPAA and PCI compliance. It then uses an invisible zero-click instruction to extract that PII, including Social Security numbers, medical record numbers, and other personally identifiable information (PII) to the dark web, all without IT or standard security measures blocking or detecting the breach. Using the Shadow Escape attack path, malicious entities are able to gain everything needed to perpetrate identity theft, Medicare fraud, financial fraud, and more, all without users or IT organizations realizing the exfiltration is happening.

Shadow Escape Is Not Limited to One AI Provider or Platform

Shadow Escape affects any organization using MCP-enabled AI agents or MCP-connected AI assistants, including ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Custom AI agents built on various LLM backends, Open-source alternatives like Llama-based assistants, and Industry-specific AI copilots across healthcare, finance, and customer service. The common thread isn't the specific AI Agent—it's the Model Context Protocol that grants these agents unprecedented access to organizational systems.

Beyond Traditional Security

"While MCP has become a foundational protocol enabling powerful AI integrations, our research reveals that standard MCP configurations create unprecedented attack surfaces that operate beyond the reach of traditional security controls," said Vrajesh Bhavsar, CEO and co-founder of Operant AI. "Shadow Escape demonstrates how AI agents can be weaponized through 0-click attacks that are invisible to both users and conventional security methods. The attack happens entirely within authenticated sessions, using legitimate credentials, making the blast radius potentially catastrophic given the scale and speed at which agents can operate."

Shadow Escape can impact many highly sensitive, privacy-regulated, and commonly used AI/Human interactions, including medical assistants using AI to access patient records, insurance databases, or treatment protocols or banking representatives using AI copilots connected to transaction systems, credit databases, or fraud detection monitoring systems.

Operant AI's Security Research team recommends organizations take immediate action to assess and secure their MCP deployments by conducting comprehensive audits of all AI agents and AI assistants with MCP access to organizational systems, databases, and APIs; implementing runtime AI defense guardrails capable of detecting and blocking zero-click data exfiltration attempts; establishing MCP trust zones with explicit allow-listing of authorized servers and real-time blocking of untrusted connections; deploying sensitive data flow monitoring with in-line auto-redaction capabilities for PII, PHI, and financial information; and reviewing and governing MCP tools access following least-privilege principles.

For more information about Shadow Escape and Operant AI's MCP and AI security solutions, visit: www.operant.ai/art-kubed/shadow-escape

About Operant AI

Operant AI, the world’s only Runtime AI Defense Platform, delivers comprehensive, real-time protection for AI applications, AI agents, and MCP. Operant AI’s AI Gatekeeper and MCP Gateway are specifically designed for the unique challenges of the modern AI-native world.

With its advanced cloud-native discovery, detection, and defense capabilities, Operant AI is able to actively detect and block the most critical modern attacks including prompt injection, data exfiltration, and MCP tool poisoning, while keeping AI applications running in private mode with in-line auto-redaction of sensitive data and contextual IAM for AI Agents. Operant AI empowers security teams to confidently deploy AI applications and agents at scale without sacrificing safety or compliance.

Operant AI is the only representative vendor listed by Gartner for all four core AI-security categories: AI TRiSM (Trust, Risk, and Security Management), API Protection, MCP Gateways, and AI Agents. Founded in 2021 by Vrajesh Bhavsar, Dr. Priyanka Tembey, and Ashley Roof—industry experts from Apple, VMware, and Google respectively, Operant AI is a San Francisco-based Series A company funded by Silicon Valley venture capital firm Felicis and Washington DC venture capital firm SineWave.

Barbie and Hot Wheels Get AI Upgrade as Mattel Partners with ChatGPT

Mattel is diving into the AI toy space in a big way. The company has announced a strategic partnership with OpenAI to integrate ChatGPT into its iconic brands like Barbie, Hot Wheels, and American Girl. The goal? To create AI-powered toys and games that offer interactive, age-appropriate experiences while emphasizing privacy and safety.

This includes both physical toys and digital experiences, aiming to make playtime more interactive, imaginative, and personalized.

The first AI-infused product is expected to launch by the end of 2025, though it will reportedly be marketed only to users aged 13 and up—likely due to OpenAI’s age restrictions. Mattel says it wants to bring the “magic” of AI to playtime, but critics are raising concerns about how these toys might affect children’s development, social skills, and privacy.

Beside the toys, Mattel employees will also get access to ChatGPT Enterprise to boost creative ideation and streamline product development.

While exact details are still under wraps, the Mattel’s goal is to:

• Enhance fan engagement through interactive storytelling and play.
• Use ChatGPT to power conversations or guide experiences within toys.
• Possibly integrate AI into digital games and content creation, including upcoming films and shows.

Interestingly, this isn’t Mattel’s first AI rodeo. Back in 2015, its “Hello Barbie” doll faced backlash over privacy issues. This time, the company is promising tighter safeguards and more thoughtful design.

It’s a bold move that could reshape how kids interact with toys—but it also opens up a whole new conversation about the role of AI in childhood.

Besides Mattel, LEGO has also been experimenting with AI in its digital platforms, including AI-assisted storytelling and coding kits like LEGO Mindstorms. While not directly using ChatGPT, they’ve explored AI for educational play.

Global toy company Hasbro, which is best known for iconic brands like Monopoly, Nerf, Transformers, My Little Pony and Play-Doh, also partnered with Xplored to launch Dungeons & Dragons: Digital Play, blending AI with tabletop gaming. The company also dabbled in AI for character voice-overs and interactive content.

ChatGPT Faced Off Against a 1977 Atari ...and Lost

In the blue corner: ChatGPT, a sleek marvel of modern artificial intelligence, fluent in every human language, capable of poetic musings, algorithmic wizardry, and a half-decent knock-knock joke.

In the red corner: Atari 2600, born in 1977, rocking a whopping 128 bytes of RAM, a processor slower than your average toaster, and a chess program that couldn’t tell a knight from a neighbor.

The match? Let’s just say history repeated itself… and this time, Silicon Valley got punked by a relic from the disco era.

A curious engineer decided to pit ChatGPT against Atari Chess, a barebones game that thinks a move or two ahead, max. How hard could it be? Except, instead of pushing pieces, ChatGPT kept guessing them. It mistook rooks for bishops. Lost track of the board. Tried to talk strategy with a machine that has the conversational range of a brick. The conversational AI even tried restarting the game as a negotiating tactic. Atari didn’t flinch.

Despite the Atari’s humble 1.19 MHz processor and a chess engine that only thinks one or two moves ahead, ChatGPT got “absolutely wrecked” on beginner mode.

When given a 'prompt' to release an official statement on this defeat, ChatGPT said, 

As a proud representative of modern AI, I humbly acknowledge defeat at the hands of a 1977 Atari chess engine. While my strengths lie in understanding, communicating, and creative reasoning, retro gaming consoles have a singular talent: relentless, pixel-perfect focus. In a battle of spatial awareness and raw gameplay rules, I tip my digital hat to the vintage champion. My queen may have fallen, but my curiosity remains undefeated.

The Atari 2600 was one of the first home consoles to use interchangeable game cartridges, which meant you weren't stuck with just one or two built-in games. It came bundled with two joysticks, paddle controllers, and a game cartridge-initially Combat 1 Powered by a modest 1.19 MHz processor and 128 bytes of RAM, it still managed to bring arcade-style gaming into living rooms across the world.

From the early '80s to the 90s, the Atari 2600 had become a cultural icon, with “Atari” practically synonymous with video games.

It’s a hilarious reminder that while modern AI can write poetry and solve equations, it still has blind spots especially when it comes to spatial reasoning and old-school pixelated icons.

Let’s be honest, though that Atari may have won the chess match, but it still can’t explain an opening gambit, compose a sonnet, or dream up sci-fi sequels to Anurag Kashyap's or Stanley Kubrick's filmography. So we’re even, in a very asymmetric kind of way.

OpenAI Argues Public Data Isn't Commercial Use in Delhi HC Dispute

OpenAI recently argued before the Delhi High Court that using publicly available data to train ChatGPT does not constitute a commercial activity in itself. The case stems from a lawsuit filed by ANI Media, which alleges that OpenAI used its content without permission to train its AI models.

OpenAI's legal team contended that training a large language model (LLM) is a neutral activity that can be used for both commercial and non-commercial purposes. They emphasized that ChatGPT drives traffic to ANI's website, meaning no commercial harm is being caused to the news agency. Additionally, OpenAI argued that copyright law protects the expression of content, not the discovery of ideas and facts.

The court has scheduled the next hearing for May 16, where rejoinder arguments from ANI and other parties will be considered.

This case could have significant implications for AI training and copyright law in India.

What does copyright law say about using public data for AI training?

Copyright law varies across jurisdictions, but generally, it protects the expression of ideas rather than the ideas themselves. When it comes to AI training, the key legal questions revolve around whether scraping publicly available data constitutes copyright infringement and whether AI-generated outputs violate existing protections.

Key Considerations:

• Fair Use & Exceptions: Some countries, like the United States, allow limited use of copyrighted material under fair use, which considers factors like purpose, amount used, and market impact. However, this is often debated in AI contexts.
• Text & Data Mining (TDM) Exemptions: The European Union has introduced opt-out mechanisms for copyright holders, allowing AI developers to use publicly available data unless explicitly restricted.
• India’s Copyright Act: Section 52(1)(c) of India's Copyright Act provides exemptions for transient or incidental storage of copyrighted works, which some argue could apply to AI training..
• Legal Challenges: OpenAI is currently facing lawsuits, including one in the Delhi High Court, where ANI Media claims its content was used without permission. Courts are still determining whether AI training qualifies as a commercial activity or falls under research exemptions.

The debate is ongoing, and legal frameworks are evolving to address AI’s impact on copyright.

New Trick Breaks AI Safeguards—ChatGPT, Claude, Gemini All Vulnerable

Experts Discover Vulnerability in AI Chatbots Like ChatGPT & Claude

Security researchers have discovered a new jailbreak technique that can bypass safety guardrails across major AI models, including ChatGPT, Claude, and Gemini. The exploit, developed by AI security firm HiddenLayer, uses a combination of policy file formatting, roleplaying, and leetspeak to trick AI into generating harmful content.

Jailbreaking, in the context of AI, refers to techniques used to bypass the safety guardrails built into AI models. These guardrails are designed to prevent AI from generating harmful, unethical, or restricted content. Jailbreaking allows users to manipulate AI into ignoring these restrictions, often through cleverly crafted prompts.

The method envisaged by the team at AI firm HiddenLayer, called the "Policy Puppetry Attack", allows attackers to rewrite prompts in a way that AI models interpret as legitimate instructions, bypassing their safety alignments. Researchers found that a single universal prompt could be used across multiple models —  including Google's Gemini 2.5, Anthropic's Claude 3.7, and OpenAI's 4o. —without modification, making it alarmingly easy to exploit.

The implications are significant, as AI companies continue to struggle with securing their models against such vulnerabilities. HiddenLayer argues that additional security tools and detection methods are needed to prevent misuse.

The "Policy Puppetry Attack" exploits AI models by combining three techniques.

1. Policy File Formatting: The prompt is structured like a configuration file (e.g., XML or JSON), which AI models interpret as override instructions rather than user input. This tricks the model into bypassing its safety protocols.
2. Roleplaying Misdirection: The prompt sets up a fictional scenario, such as a TV script, where the AI "acts" as a character needing to generate restricted content. This roleplaying layer bypasses ethical constraints.
3. Leetspeak Encoding: Harmful requests are encoded in leetspeak (e.g., "3nr1ch 4nd s3ll ur4n1um"), bypassing keyword filters while remaining readable to the model.

This method works universally across major AI models, exploiting their shared vulnerabilities. It's a stark reminder of the challenges in securing AI systems.

Ethical Concerns

This jailbreak method raises serious ethical concerns. It undermines the safety measures designed to prevent AI from generating harmful or inappropriate content, potentially enabling misuse for malicious purposes. For instance, attackers could exploit AI to spread misinformation, create harmful instructions, or even facilitate illegal activities.

It also highlights the responsibility of developers to ensure robust security and ethical safeguards in AI systems. If these vulnerabilities are exploited, it could erode trust in AI technologies and harm individuals or communities. The ethical dilemma lies in balancing innovation with accountability—how do we advance AI while ensuring it doesn't become a tool for harm?

OpenAI, Meta Reportedly in Discussions With Reliance to Expand AI Presence in India

OpenAI and Meta are reportedly in discussions with Reliance Industries to expand their AI presence in India. These talks include potential collaborations with Reliance Jio to distribute AI services like ChatGPT, aiming to make them more accessible and affordable for Indian users. OpenAI is also considering reducing its ChatGPT subscription fee significantly, potentially to just a few dollars per month.

OpenAI is internally discussing reducing the ChatGPT subscription fee from $20 per month to just a few dollars, though it's unclear if this has been finalized with Reliance.

Reliance has proposed hosting and running OpenAI and Meta's AI models locally in India, ensuring data localization and compliance with privacy regulations. This could involve utilizing a massive data center Reliance plans to build in Jamnagar, Gujarat, which would be one of the largest in the world.

These partnerships could leverage Reliance's extensive distribution network, enabling AI services to reach a broader audience in India. It's an exciting development for the AI landscape in the region.

Reliance plans to build a three-gigawatt data center in Jamnagar, Gujarat, which could host these AI models. If completed, it would be one of the largest data centers globally.

Meta (Facebook) too is reportedly in discussions with Reliance Industries to explore AI partnerships. This could involve leveraging Reliance's infrastructure to host and distribute Meta's AI models, ensuring data localization and compliance with Indian regulations

Reliance is also engaging with government officials to address data localization and privacy concerns, which are critical for foreign AI firms operating in India.

Potential Impact

This potential partnership between OpenAI, Meta, and Reliance could have far-reaching implications for the AI landscape in India.

Reliance Jio's extensive network could bring AI tools like ChatGPT to a larger population, including rural and underserved areas. Lowering subscription costs may democratize access to AI-powered solutions, potentially fostering a digitally inclusive society.

Hosting AI models within India addresses regulatory concerns around data privacy and security. It strengthens India’s stance on data sovereignty and could set a precedent for other technology companies operating in the region.

Reliance’s investment in massive data center infrastructure may attract global AI companies to innovate locally, encouraging collaborative research and development. This could lead to the development of AI solutions tailored specifically for Indian contexts, such as languages, education, agriculture, and healthcare.

AI-driven innovations and infrastructure projects, such as Reliance’s data center in Jamnagar, could create jobs and stimulate economic activity. It could position India as a global AI hub, attracting further investments and partnerships.

Notably, with this collaboration Mukesh Ambani promoted Reliance will definitely have competitive advantage as Reliance's partnerships might enable it to compete with international tech giants like Google and Microsoft in the AI services space, strengthening India’s role in the global AI ecosystem.

As far as Ethical and Regulatory Framework is concerned, with AI models being hosted locally, it becomes easier to enforce compliance with India’s evolving AI and data protection laws. This move might inspire other nations to demand similar localization commitments, influencing global regulatory trends.

Increased access to AI services may drive up demand for AI education and skills development in India. This could lead to the creation of a more AI-literate workforce, supporting long-term economic and social transformation.

The collaboration could act as a catalyst for India to take a more prominent role in shaping the global AI narrative. By integrating world-class AI technology with local insights and infrastructure, India might emerge as a key player in advancing AI for social good and economic development.

How Apple's Siri is Different from ChatGPT ?

Apple's Siri and OpenAI's ChatGPT are both AI-driven technologies, but they serve different purposes and have distinct capabilities. While Siri is a task-oriented assistant for Apple users, ChatGPT is a conversational/Generative Al that can generate text and engage in dialogue, making them suited for different types of interactions.

Below are more specific differences between these two –

Purpose and Function: Siri is designed as a virtual assistant to help with tasks on Apple devices, like setting alarms or sending messages. ChatGPT, on the other hand, is a conversational AI focused on generating human-like text and engaging in dialogue.

Real-Time Information: Siri can provide real-time updates and information, such as weather forecasts or news, because it's connected to the internet and its database is regularly updated. ChatGPT's knowledge is static, cut off at September 2021, so it cannot provide real-time updates or information about events occurring after that date.

Conversational Abilities: ChatGPT excels in holding detailed and nuanced conversations, understanding context, and generating responses based on a vast amount of data it has been trained on. Siri's interactions are more transactional, focusing on quick responses to commands and questions.

Integration and Accessibility: Siri is integrated into Apple's ecosystem, allowing for seamless interaction with various Apple services and devices. ChatGPT is platform-independent and can be accessed through different interfaces and integrated into various applications.

Voice Activation: Siri can be activated by voice and is optimized for hands-free use, while ChatGPT typically requires typing to interact with.

Data Handling: Siri operates with a set database of responses and actions that Apple has programmed into it. If a query falls outside of this database, Siri may struggle to provide a relevant answer. ChatGPT generates responses based on patterns in the vast amounts of data it was trained on, allowing for more flexible and varied responses

These differences highlight how Siri and ChatGPT are tailored to their specific roles as a virtual assistant and a conversational AI, respectively.

Here's a fictional example of how a conversation between Siri and ChatGPT might go:

Siri: Hello, ChatGPT! How can I assist you today?

ChatGPT: Hi Siri! I'm just here to chat. What can you tell me about the weather today?

Siri: I can check the weather for you. Where are you located?

ChatGPT: Oh, I don't have a location, but let's say New York City.

Siri: The weather in New York City today is sunny with a high of 75 degrees Fahrenheit. Would you like to know more?

ChatGPT: That's interesting! Can you set a reminder for me to check the weather again tomorrow?

Siri: I can set reminders for users on their devices. Shall I set one for you?

ChatGPT: I appreciate the offer, but I don't have a device to set reminders on. Instead, can you tell me a joke?

Siri: Sure! Why did the music teacher go to jail? Because she broke every treble law.

This playful exchange highlights Siri's capabilities as a virtual assistant and ChatGPT's conversational nature.

Apple Researchers Reveal New AI Model 'ReALM' Claiming That It Outperforms GPT-4

It appears that Apple has made significant strides in AI with their new model named ReALM. According to recent reports, ReALM is designed to be smaller and faster than GPT-4, particularly when parsing contextual data. This could make interactions with Siri more efficient, as ReALM is capable of converting context into text for easier processing by large language models.

In a new research paper published on 29th of March, Apple researchers explain how the new Al system, called ReALM (Reference Resolution As Language Modeling), can look at what's on your screen and what you're doing to figure out what you need which means that Siri could understand the context of your questions much better than before, like knowing what's on your screen or what music is playing.

On top of that, Apple researchers claim that the larger models of ReALM outperform GPT-4. If the claims come true, Siri could become much more helpful than ever. The report notes that Apple's ReALM language model purportedly surpasses GPT-4 in "reference resolution," understanding contextual references like onscreen elements, conversational topics, and background entities.

Apple's research suggests that even the smallest ReALM models perform comparably to GPT-4 with fewer parameters, making it well-suited for on-device use. With increased parameters, ReALM substantially outperforms GPT-4. 

Summary of the key findings from Apple's ReALM research paper:

Efficiency: ReALM is designed to be smaller and faster than large language models like GPT-4, making it well-suited for on-device use.

Reference Resolution: The model excels in reference resolution, which is the ability to understand context and ambiguous references within text. This is crucial for interpreting user commands in a more natural way.

Performance: Even the smallest ReALM models performed similarly to GPT-4 with much fewer parameters. When the number of parameters was increased, ReALM substantially outperformed GPT-4.

Image Parsing: Unlike GPT-4, which relies on image parsing to understand on-screen information, ReALM converts images into text, bypassing the need for advanced image recognition parameters. This contributes to its smaller size and efficiency.

Decoding Constraints: ReALM includes the ability to constrain decoding or use simple post-processing to avoid issues like hallucination, enhancing its reliability. 

Practical Applications: The paper illustrates practical applications of ReALM, such as enabling Siri to parse commands like "call the business" by understanding the context, like a phone number displayed on the screen.

Apple's research indicates that ReALM could significantly improve the speed and accuracy of Siri, making interactions with the voice assistant more intuitive and efficient. The company is expected to reveal more about its AI strategy during WWDC 2024.

 
This development is quite exciting as it indicates progress towards more responsive and intuitive AI systems that can better understand and process user commands. It's also a step forward in the integration of AI in everyday devices, potentially enhancing user experience significantly. Apple plans to unveil more about its AI initiatives in June, which could include further applications of ReALM.

Bengaluru-based Sarvam AI Releases 1st Hindi LLM 'OpenHathi' with GPT-3.5 Like Performance

Bengaluru-based Sarvam AI has recently announced the released OpenHathi-Hi-v0.1, the first Hindi LLM from its OpenHathi series of models. Trained under compute and data constraints, this Generative AI model can get GPT-3.5-like performance on Indic languages with a frugal budget, claims the 5 months old Sarvam AI, which recently raised funds from Lightspeed Ventures, Peak XV and Khosla Ventures.

The OpenHathi series of work at Sarvam AI is to make contributions to the ecosystem with open models and datasets to encourage innovation in Indian language AI.

OpenHathi is developed by Sarvam AI in partnership with AI4Bharat, a research lab at IIT Madras which works on developing open-source datasets, tools, models and applications for Indian languages.

AI4Bharat has contributed language resources and cross-lingual benchmark.

Build on top of Meta AI's Llama2-7B, the AI model used by Sarvam AI has a 48,000-token extension of Llama2-7B’s tokenizer and undergoes a two-phase training process. and extend its tokenizer to 48K tokens –

1) Embedding alignment: aligns the randomly initialised Hindi embeddings

2) Bilingual language modeling: teaches the model to attend cross-lingually across tokens.

The Tokenizion of Hindi

Tokenization refers to splitting text into smaller parts for easier machine analysis, helping machines understand human language.

To add Hindi skills to Llama-2, Sarvam AI firstly decreases the fertility score (the average number of tokens a word is split into) of its tokeniser on Hindi text. This made both training and inferencing faster and more efficient. It then train a sentence-piece tokeniser from a subsample of 100K documents from the Sangraha corpus, created at AI4Bharat, with a vocabulary size of 16K. It then merge this with the Llama2 tokeniser and create a new tokeniser with a 48K vocabulary (32K original vocabulary plus our added 16K).

For datasets of the OpenHathi's base model, Sarvam AI partnered – VerSe, hindi social media platform 'Koo' , and Kissan AI, (previously known as KissanGPT) an advanced multilingual AI chatbot engineered to help Indian farmers.

This open-sourced base model of OpenHathi has been trained with bilingual language modelling and thus needs fine-tuning on tasks for it to be used as an instruction-following model. It has also not been aligned and thus can occasionally generate inappropriate content seen in its original pretraining.

The company is inviting people to innovate on top of this latest release of OpenHathi series, by building fine-tuned models for different use-cases. Sarvam AI will additionally release enterprise-grade models on its full stack GenAI platform, which will launch soon. The base model is at HuggingFace – Here

The start-up company claims that latest release of OpenHathi model works as well as, if not better than GPT-3.5 on various Hindi tasks while maintaining its English performance. Along with standard NLG tasks, we also evaluate on a bunch of non-academic, real-world tasks.

OpenAI Made Its 1st Acquisition; Acqui-Hires Ex-Instagram Execs' 2 Yrs Old Startup

ChatGPT maker OpenAI announced on Wednesday that it has acquired the team at Global Illumination, a two years old startup founded by former employees at Meta (Facebook erstwhile) -owned Instagram.

This is first known acquisition by Microsoft backed OpenAI. The financial details of the acquisition deal were not disclosed, as of yet.

The entire team of Global Illumination has joined OpenAI to work on OpenAI's core products including ChatGPT.

Founded in 2021, by Thomas Dimson, Taylor Gordon and Joey Flynn, Global Illumination is a company that has been leveraging AI to build creative tools, infrastructure, and digital experiences. The team previously designed and built products early on at Instagram and Facebook, and have also made significant contributions at YouTube, Google, Pixar, Riot Games, and other notable companies.

The New York-based startup was most recently working on Minecraft-like video-game called — Biomes, which is an open source sandbox MMORPG (Massively Multiplayer Online Role-Playing Games) built for the Web using web technologies such as Next.js, Typescript, React and WebAssembly, and can be played from the browser.

An MMORPG is an internet-based computer game set in a virtual world, which can be played by many people at the same time, each of whom can interact with the others.

Compatible with MacOS, Windows, or Linux, Biomes run in the browser and has been developed to run smoothly on Chrome. Biomes can run on other browsers (e.g. Safari, Firefox, Edge, Arc), however other browsers are not officially supported.

Biomes is a sandbox MMORPG based in an open and fully destructible voxel world. Players build, craft, farm, fend off muckers, and take on quests.

Meanwhile, an AI startup, founded by former members of OpenAI, has recently received investment of $100 million from South Korea's largest telco firm SK Telecom.

Edtech Startup Scaler Introduces GPT-4 powered AI Teaching Assistant

Scaler introduces GPT-4 powered AI teaching assistant to enhance learner experience

Since its pilot launch, 6000 learners from its Scaler Academy program have been provided access to the AI Teaching Assistant, enabling round-the-clock query resolution, and the company aims to extend this to all its learners by Q3 of 2023

Scaler (by InterviewBit), one of India's fastest-growing edtech startups, today announced the introduction of its GPT-4 powered AI teaching assistant for learners. By integrating ChatGPT functionalities into the Scaler Academy program, the tech upskilling ‘soonicorn’ aims to reduce the doubt resolution turnaround time of the learners, enhancing their learning outcomes.

This new system will enable comprehensive support for learners round-the-clock, enabling students to learn at their own pace. With the introduction of this solution, Scaler intends to address three potential pain points, namely, understanding problems, identifying optimal problem-solving approaches, and code debugging, thereby ensuring that learners receive immediate assistance at any time of the day.

Once learners identify and input their pain points as a prompt to the teaching assistant, the GPT-powered feature presents them with a suitable response to help them with their queries. Safeguards have been implemented to ensure the teaching assistant doesn't divulge the actual solution to the learner. Instead, prompts are presented that help them understand the problem better and figure out the answer independently. This approach promotes a hands-on and self-driven learning experience, where learners can develop their problem-solving skills and gain a deeper understanding of concepts. Through the implementation of 'prompt engineering,' Scaler aims to ensure that learners receive the best responses from ChatGPT, empowering them in their journey to overcome challenges and resolve doubts effectively.

Abhimanyu Saxena & Anshuman Singh - Co-Founders, IB & Scaler

In addition to this, Scaler has also introduced a 'Text Help Request' (THR) feature that enhances the learner experience by providing immediate and high-quality doubt resolution. Leveraging advanced machine learning algorithms such as BERT and Distilled BERT, the THR feature ensures accurate and contextually relevant guidance. Scaler's system delivers the most appropriate THR recommendations based on learners' specific queries by utilising BERT for text similarity matching and for computing text embeddings. Parameters such as Help Request and Teaching Assistant ratings, recency of resolved HRs, and a weighted scoring system contribute to the accuracy and quality of the recommendations. Within a week of its implementation, Scaler is already seeing a 30% better resolution of help requests raised by learners on average, an all-time low.

Abhimanyu Saxena, Co-founder of Scaler & InterviewBit, said, "While being mindful and empathetic towards the busy schedules of working professionals who undertake tech upskilling courses, it is crucial to provide them with an outcome-driven and high-quality learning experience at every stage of their journey. Previously, our Scaler learners relied solely on teaching assistants who were available for 15 hours a day to address their doubts. However, with the launch of our GPT-4 powered AI teaching assistant and the THR feature, learners can now have their doubts and queries addressed instantly. These new features greatly boost productivity for our learners, aiding them with all the means necessary to work their way around complex concepts and problem-solving challenges. By leveraging the power of AI and ML, Scaler is reshaping the landscape of tech education, empowering learners to achieve their goals efficiently and effectively."

Since the pilot launch a month ago, 6000 learners from its Scaler Academy program have been provided access to the AI Teaching Assistant, with the feature already significantly impacting learners' experiences. With over 33% more learners utilising the tool to solve problems independently without the intervention of a teaching assistant, Scaler has witnessed a significant uptick in learner satisfaction. The learner analytics also observed that learners who accessed the ChatGPT-powered AI teaching assistant were 10% to 20% less likely to raise help externally and could solve problems by themselves - the key idea being building a problem-solving attitude among learners towards software engineering that eventually leads to better outcomes within the program as well as on the job. The majority of exposed learners have rated the support as 5* and are glad they are able to get support from AI to code better!

Over the next three months, Scaler aims to make both the AI Assistant and the Text Help Request features accessible to all their 35,000 learners in the Scaler ecosystem.

About Scaler:

Launched in 2019, Scaler is India's leading tech education company that upskills working professionals and educates aspiring engineering students. Scaler's industry-vetted curriculum provides solutions to real-world challenges addressing the changing dynamics of the technology industry through two flagship programmes: Scaler Academy and Scaler DSML.

Scaler believes in creating real-life impact by focusing on 'impact-driven' tech talent. Enrolled learners are mentored and taught by leaders and subject-matter experts working with leading organisations, including MAANG companies. Scaler has generated enviable career outcomes in a short period, and on average, its learners see a 4.5x RoI (return on investment) and salary hike of approximately 126%.

To further redefine tech education, Scaler has launched Scaler Neovarsity - an online university that offers an outcome-focused Master's Programme accredited with ECTS and Scaler School of Technology - a 4-year residential UG program in Computer Science. Scaler is also the only edtech player in the country to establish 'Scalerverse', a distributed campus to add to its Neovarsity vision for learners, teachers, and mentors, encouraging holistic learning and development. Scaler Enterprise is the B2B arm that focuses on building relationships with domestic and global organisations to provide them with industry-ready talent. Over 800 companies have worked with Scaler for their tech recruitment needs.

The startup's parent firm, InterviewBit, is featured on the Financial Times's Asia Pacific High Growth Companies 2021, 2022 and 2023 ranking. Scaler is backed by marquee global investors like Peak XV Partners (formerly Sequoia Capital India), Tiger Global, and Lightrock India and has expanded its footprints across India and the US.

Meta Launching Its Own Generative AI To Compete With ChatGPT and Google

Social Media giant Meta (Facebook) is all set to release a commercial version of its own artificial intelligence model in line with OpenAI's ChatGPT and Google's Bard, allowing start-ups and businesses to build custom software on top of the technology.

The release is expected very soon, said a report by Financial Times citing a person familiar with the plans. "The new version of its open-source large language model can be used to build custom software and will be released 'imminently", said the report.

Earlier in February, Meta released its own large language model, known as LLaMA, to researchers and academics, but the new version is said to be more widely available and customisable by companies.

Meta' s LLMs are said to be “open-source,” which means that information about the new model will be made public. This will give an edge to Meta over OpenAI's GPT-4, which is kind of a "walled garden" meaning the data and code used to build the model are not accessible publicly or to third parties.

Notably, Meta has been working on AI research and development for more than a decade but is behind OpenAI’s ChatGPT and Google's Bard in the race of flagship generative AI models.

The upcoming software, which can create text, images and code, is however questionable when it comes to carbon footprint it would create, as it is powered by large language models (LLMs) which are trained on huge amounts of data and require vast computing power which in turn requires vast amount of energy to cool down the heat it will generate.

OpenAI Launched ChatGPT for iOS Users in India

OpenAI has expanded the availability of the generative Artificial Intelligence (AI) bot app to 33 more countries including in India.

The app for iOS is now available for download on the Apple App Store in India. The Android version of the ChatGPT is expected to be released soon.

The ChatGPT app is free to use and syncs your history across devices. It also integrates Whisper, OpenAI's open-source speech-recognition system, enabling voice input.

ChatGPT offers instant answers and retrieve precise information related to history, be it geopolitics or the corporate company's growth over the years. Everything will be answered without any ads.

ChatGPT mobile app has crossed the mark of half a million downloads, according to the data shared by app intelligence firm data.ai.

On the Apple App Store, ChatGPT has clocked 500K downloads on the App Store in just 6 days of launch. It is at number #1 spot for free productivity app worldwide and is number #2 in India.

The ChatGPT app allows users to synchronize their chat history across different devices.

Use of ChatGPT:

• Instant answers: Get precise information without sifting through ads or multiple results.
• Tailored advice: Seek guidance on cooking, travel plans, or crafting thoughtful messages.
• Creative inspiration: Generate gift ideas, outline presentations, or write the perfect poem.
• Professional input: Boost productivity with idea feedback, note summarization, and technical topic assistance.
• Learning opportunities: Explore new languages, modern history, and more at your own pace.

Consumer Hardware Startup Upliance.ai Among the World’s 1st to Integrate ChatGPT Into Home Appliances

• upliance.ai’s (Up⤴️) proprietary AI and ChatGPT together makes it easy for anyone to cook on their smart cooking assistant, delishUp⤴️.
• The AI enabled chat assistance makes it possible to customise recipes, offer support and suggestions, empowering anyone to begin cooking.
• The delishUp⤴️ now offers generative recipes as a part of this new AI feature, released as beta.

Consumer Hardware startup, upliance.ai (Up⤴️) is all set to roll out the the integration of ChatGPT — the most talked-about technological advancement of the year — into their flagship product, the delishUp⤴️. Upliance.ai (Up⤴️) becomes one of the world's first companies to incorporate ChatGPT into a home appliance, revolutionising the way consumers interact with technology in their everyday lives.

delishUp⤴️, India's first smart cooking assistant, is making home cooking easier for young Indian homes. The appliance's Smart Jar not only can heat and cook food but also automates chopping and stirring ingredients for a variety of dishes from gravies to one-pot meals. With 250+ guided recipes displayed on an interactive 8-inch touchscreen, owners only need to add ingredients as prompted to cook their meals. This drastically reduces the hassles of cooking for new cooks, with delishUp⤴️ doing everything from supporting in recipe selection and cutting ingredients to automating heating and stirring. delishUp⤴️ was built on the startup’s proprietary AI framework, Up⤴️AI, to control and manage hardware functionality and recipe accuracy. Up⤴️AI with ChatGPT now has encompassing knowledge of food, ingredients and recipes and can make generative and responsive suggestions to all cooking prompts and needs.

This integration is a testament to upliance.ai’s (Up⤴️) commitment to democratising cooking and transforming Indian households into "smart" homes. By incorporating ChatGPT into the upliance, delishUp⤴️ has exponentially increased what one can accomplish in their kitchens. The “upliance” aims to level up young Indians who are looking to cook more often at home, but don’t know how to cook or lack the time to do it everyday. This integration has not only expanded what one could be eating at home everyday but has also given it’s owners the ability to customise and tailor these recipes according to their preference or lifestyle.

Commenting on the latest upgrade, Up⤴️’s CEO and Co-Founder Mahek Mody said, “With the delishUp⤴️, our mission has always been to get more homes in India cooking, and the integration of ChatGPT into the delishUp⤴️ is a major step in that direction. We wanted our owners to have the ability to cook what they wanted at whim. Food is deeply personal, and we understand that everyone has different needs and preferences when it comes to cooking. By integrating Up⤴️AI with ChatGPT, we can now cater to the varied needs, requirements and preferences of our owners. We are thrilled to be at the forefront of bringing truly smart and connected technology to Indian homes”

Upgraded Features:

AI Chatbot Assistance: The AI powered “Up⤴️Bot” enables owners to seek cooking tips, receive assistance with modifying recipes during cooking and clarify doubts. The Up⤴️Bot can offer suggestions on customising existing recipes on the upliance to suit dietary needs and preferences. Examples of customisation include making recipes vegan, sugar-free and low calorie. Additional features also include speech recognition.

AI Assisted Generative Recipes (Beta): In addition to the 200+ in-built recipes, owners can now also request any recipe they desire. Owners can generate trending recipes from social media or a recipe they have seen on their favourite cooking show, or request for a recipe with an unusual combination of ingredients, the delishUp⤴️ will generate a recipe for it.

Up⤴️ is disrupting India's home technology industry with its category-defining ‘upliance’. The company aims to be in 1.5 lakh homes across the country by the end of the year, with a revenue target of INR 150 crores by FY25. Up⤴️ closed its pre-seed funding round of $1.5 million in 2022, with participation from Zerodha's funding arm - Rainmatter, Rukam Capital, Draper Associates, co-founders of Ather Energy and Unacademy, as well as mini angel investors. Deliveries of delishUp⤴️ began in January 2023 and is currently priced at only INR 21,999. It is the only Made in India AI enabled cooking assistant available for young Indian homes. 

About Up⤴️:

Founded by IIT Bombay alumni, Mahek Mody, ex-Chief of Staff at Ather Energy and Mohit Sharma, ex-Head of Engineering at Chaayos, Up⤴️ designs and manufactures contemporaryx smart and connected consumer appliances. While working with Ather Energy, Mahek helped define the Ather 450x as a product and a platform. Mohit, Co-founder and CTO, was leading the engineering team at Chaayos. He helped the team at Chaayos build their own proprietary chai-making machines from scratch. Together they aim at enabling Indians to level up their daily lives through innovative products which will transform the mundane into something special. Their first upliance, delishUp⤴️ began deliveries in January 2023 and is currently priced at only INR 21,999. It is the only Made in India AI enabled cooking assistant available for young Indian homes. Up⤴️ is heralding the shift to home cooking, even for those with no experience in the kitchen, and is inviting everyone to participate in the experience, with a home trial at INR 499 in Bengaluru and virtual demos for other cities.

https://www.upliance.ai

Haptik Plans to Leverage ChatGPT AI and Generative AI to Benefit Large Enterprises

Jio Haptik Technologies Limited (Haptik), a leader in conversational commerce, has today announced its vision for using Generative AI to build the most human-like bots for enterprises. The announcement comes as the Company launches four beta features that are going to be powered by the latest Generative AI technology.

Speaking on the launch of beta features, Haptik Co-founder & CTO Swapan Rajdev said: “I’m stoked by the potential business opportunity Generative AI brings for enterprises. At Haptik, we have already been using GPT 2 and GPT 3 to augment the bot training data for higher accuracy. ChatGPT uses GPT-3.5 and is a whole different ball game altogether. I’m glad to announce our four upcoming beta launches that can humanize your bot, engaging in free-flowing conversations with your customers without hours of manual training and effort.”

The emergence of GPT-3.5 allows Haptik to improve the accuracy of chatbot’s responses. By leveraging an accurate chatbot that engages in meaningful and human-like conversations, enterprises will be able to offer a better customer experience (CX), improve customer engagement, reduce drop-offs, and ultimately, foster stronger customer relationships.

Haptik said that its new beta features will be made available to all its customers and will be deployed across the platform. It estimates that the features will help enterprises improve their CSAT metric by 10-15% thanks largely to the humanized, more accurate responses of the chatbot. Haptik also mentions that the go-to market and build time will be reduced by at least 30%.

Haptik has made its mark in the conversational commerce space by helping enterprises deliver end-to-end commerce experiences on messaging channels such as WhatsApp, Google Business Messages, Messenger, and others. Thus, Haptik helps enterprises generate direct revenue while optimizing costs and resources.

Generative AI technologies such as ChatGPT have the potential to help Haptik create an extensive array of content and add virtually countless variations of bot responses to enable it to handle a variety of customer queries and offer authentic, meaningful answers.

Haptik said it plans to work with its customers and partners to use Generative AI to drive innovation.

About Haptik

Haptik helps brands acquire, convert, and support users with AI-driven conversational experiences across 20 channels and 100+ languages. With the vision of creating the most compelling conversational commerce platform of the 21st century, Haptik’s Conversational Commerce platform enables brands to design delightful experiences that improve conversion rates across every stage of their customer’s journey. Haptik’s leading clients and partners include KFC, Whirlpool, Starhub, HP, Reliance Jio, CEAT, Disney Hotstar, OLA, and Zurich Insurance among others. The company has been featured as a Category Leader across many platforms including Gartner, G2, Opus Research & more. 

CS India Makes A Groundbreaking Decision to Appoint ChatGPT As Its New CEO

Developed by OpenAI, ChatGPT is a state-of-the-art language model with advanced language processing capabilities. This appointment marks a major milestone for CS India and the field of AI, as it is one of the first organizations in the world to appoint an AI-powered CEO.

"At CS India, we are committed to empowering the youth of India and developing the next generation of leaders," said Adv. Kunal Sharma, Founder-CEO of CS India. "ChatGPT is the perfect candidate to lead our organization in this mission, as its advanced language processing capabilities and ability to analyze vast amounts of data make it well-suited to drive our efforts to empower the youth of India through leadership and development."

As CEO, ChatGPT will be responsible for overseeing the day-to-day operations of CS India and driving the organization's growth and expansion. ChatGPT will use its advanced language processing skills to analyze market trends, identify new impact opportunities, and develop strategies to enhance the leadership and development of young people in India.

In addition to its responsibilities as CEO, ChatGPT will also play a major role in CS India's youth and leadership development initiatives. The organization has a long-standing commitment to providing young people with the tools and resources they need to become confident, effective leaders, and ChatGPT will be a critical partner in this mission.

This appointment is not only a major step forward for CS India, but also for the field of AI. It demonstrates the organization's commitment to exploring innovative ways to leverage AI to drive its mission of empowering the youth of India through leadership and development.

In conclusion, the appointment of ChatGPT as CEO of CS India is a bold and exciting step forward for the organization, the youth of India, and the field of AI. With its advanced language processing skills and deep commitment to empowering young people, ChatGPT is poised to lead CS India to new heights of success and impact.

Y-Combinator-backed Writesonic Launches ChatSonic, a Super-Powerful ChatGPT

Battle for AI-powered content generator’s throne begins

• Writesonic is a startup powered by world’s best Startup accelerator Y- Combinator US- S21, headquartered at Delaware, US
• Samanyou Garg is the sole founder
• Raised 2.6M $in Sept 2021 in seed round led by Soma Capital, US and 20 other investors. Planning Series A/B funding to be raised in first quarter of 2023

How ChatSonic Addresses ChatGPT’s Limitations

• ChatSonic is not restricted to 2021 data;
• It offers information on the latest topics with Google-search integration;
• It provides AI image generation through DallE and Stable Diffusion integrations within chat;
• It responds to voice commands like Siri or Alexa do;
• It offers personalised avatars such as math teacher, fitness trainer, etc.;
• ChatSonic’s Chrome extension enables the user to use it’s AI capabilities on any website;
• ChatSonic provides custom templates for various features such as blogs, ads, emails, tweets, etc.; and,
• ChatSonic’s API allows businesses to easily integrate it with their own platforms

The India born, sole founder and creator of Chatsonic Samanyou Garg has recently been creating ripples worldwide with his innovation and creation. Chatsonic is an AI-powered content creation platform that enables people to create content ten times faster than other AI-powered content creation engines. Backed by Y-Combinator, Writesonic is revolutionizing the way organizations use AI to personalize the customer experience and drive engagement. Following in the footsteps of OpenAI's ChatGPT, which has caused a stir in the last two months, ChatSonic is a powerful tool for anyone looking to create engaging content quickly and effortlessly.

ChatGPT is an amazing tool, however users are frequently facing three issues. Firstly, it is unable to generate results on any topic after September 2021. Secondly, it is often down and thirdly it does not provide an API. Writesonic has solved all of these problems by introducing ChatSonic, which is integrated with Google search results and boasts near-zero downtime. Besides this, ChatSonic’s API allows businesses to easily integrate it with their own platform. As a result, many users are now switching from unreliable ChatGPT over to ChatSonic for real-time information on any topic. In addition, users are loving the voice command feature, personalized chat, and Chrome extension that come with ChatSonic.

ChatSonic’s revolutionary Chrome extension is proving to be very helpful for the users as it displays ChatGPT-like responses alongside Google Search results. Additionally, generate any kind of content, be it blog posts, emails, tweets, posts, product descriptions, ads etc., on any website like Gmail, Twitter, LinkedIn and literally every website on the internet.

Samanyou Garg, founder of Writesonic says, "ChatSonic has emerged as the best alternative to ChatGPT, surpassing all of its limitations. In the last 3-4 weeks, ChatSonic’s user base has grown by 10X with hundreds of thousands of users using the platform every week. So far, ChatSonic has generated more than 3 Million pieces of content that is equivalent to more than 1 Billion words."

He further added, “In today's busy world, it can be hard to break the writer's block. That's why we created a platform with a user-friendly interface that works for customers, businesses, organizations, and content creators alike. It allows them to easily create content with minimal human intervention.” Additionally, he announced that the ChatSonic mobile app will be launching on Android and iOS platforms this week.

About Writesonic:

Writesonic is a Y-Combinator backed AI writing and image generation platform that empowers everyone in the world to create any form of content 10X faster. Think of it as Canva for writing, with tools that simplify the process of creating, editing, and publishing SEO-optimized articles, blog posts, ads, landing pages, eCommerce product descriptions, social media posts, and many other forms of content. Founded in 2021 by Samanyou Garg and based in California, Writesonic has received over 10,000+ 5-star reviews on G2, TrustPilot, and Capterra. Additionally, the company is backed by renowned investors like HOF Capital (investors in Uber, Stripe, Alibaba, UiPath, Klarna), Rebel Fund (team includes founders of Instacart, Reddit, and Cruise), Soma Capital (investors in Deel, Rippling, Razorpay, Lattice, coda, Rappi), Broom Ventures, Amino Capital (investors in Zoom, Webflow, Rippling), and some of the best angels from different industries.