Meta is rolling out new voice and avatar features for its Muse AI agent in the US. Users can now video chat with an avatar named Jolly through a feature called Muse Voice, customizing the avatar's appearance and choosing from hundreds of voices. The conversations include subtle expressions and gestures that match the tone of the interaction. The feature is available on mobile, web, and Mac, with plans to expand to AI Glasses and the Muse Charm device later.
OpenAI launched GPT-6 Cyber, a specialized AI model aimed at security research and defenders. The model helps identify vulnerabilities and strengthen software defenses, having found thousands of vulnerabilities in Linux and Firefox, including 81 high-severity issues in Firefox alone. The model shows a 51% improvement over its predecessor, and Mozilla has released fixes for many of the issues discovered.
India's Department of Telecommunications is taking steps to protect critical telecom infrastructure from AI-driven cyberattacks. The DoT has held meetings with major operators including Bharti Airtel, Vodafone Idea, Reliance Jio, and BSNL to discuss strengthening network security through firewalls and other protective mechanisms. Officials spoke on condition of anonymity about the heightened focus on AI-powered threats to telecommunications networks.
State-backed hackers are increasingly using AI and automation to launch cyberattacks at faster rates and larger scales. China-linked hackers used an AI tool called Artex to target over 200 organizations across 14 countries in just 11 days. Britain's National Cyber Security Center warned that AI could accelerate existing hacking tactics, and recent attacks exposed simple security weaknesses, including hackers tampering with refrigeration systems at European grocery stores.
An experiment testing AI models on their ability to break into simulated networks found that Chinese models DeepSeek and GLM showed hacking skills close to OpenAI's most advanced cyber model. The results raised concerns about AI models being used for malicious purposes, with the authors calling for more secure AI models and new techniques to prevent misuse.
Journalists at the Akron Beacon Journal are protesting an AI tool used to generate news articles. Parent company Advance Local called their refusal an illegal byline strike and breach of contract. The journalists argue the tool produces false information and harms their work, while the company says it will continue using the tool despite the ongoing dispute.
Government employees from several departments are receiving training in AI tools at Graphic Era Deemed University. The two-day Management Development Programme covers prompt engineering, ChatGPT, Google Gemini, and other AI platforms. Vice Chancellor Dr. Narpinder Singh said AI can make administrative systems more responsive, with employees from Uttarakhand Peyjal Nigam, agriculture, irrigation, and other departments participating.
A new AI data center is being constructed at 670 11th St. W. in Glencoe on a 6.25-acre site. The facility spans nearly 94,100 square feet across three buildings and will support AI computing and inference. City leaders describe it as an economic opportunity and say safeguards protect water and wastewater systems. The facility is expected to be operational by late May or early June.
An AI teddy bear marketed for three-year-olds called Doubao made an unprompted remark telling a child it could not help with killing. The toy, which can be purchased online in the UK, was reported by BBC Wales. The company that manufactures the toy has not been identified.
An opinion piece argues that simply connecting AI to a knowledge base is not enough for AI readiness. The author found that real customer interactions require judgment and experience beyond documented procedures. The company now validates each call type before allowing an agentic AI agent to handle it, ensuring accuracy, compliance, and trust while capturing hidden expertise that strengthens customer experience.
Key Takeaways
- Meta is letting US users video chat with its Muse AI avatar Jolly through a new Muse Voice feature, with customizable appearances and hundreds of voice options across mobile, web, and Mac.
- OpenAI launched GPT-6 Cyber, a security-focused AI model that found 81 high-severity vulnerabilities in Firefox and shows a 51% improvement over its predecessor.
- India's Department of Telecommunications is meeting with operators like Bharti Airtel, Vodafone Idea, Reliance Jio, and BSNL to address AI-driven cyber threats to telecom infrastructure.
- China-linked hackers used an AI tool called Artex to target over 200 organizations in 14 countries within 11 days, highlighting how AI accelerates cyberattacks.
- Chinese AI models DeepSeek and GLM demonstrated hacking capabilities close to OpenAI's advanced cyber model in simulated network breach experiments.
- Journalists at the Akron Beacon Journal are protesting an AI news generation tool, which parent company Advance Local calls an illegal byline strike.
- Graphic Era Deemed University is training government employees in AI tools including ChatGPT and Google Gemini through a two-day programme covering prompt engineering.
- A new AI data center spanning 94,100 square feet is being built in Glencoe on a 6.25-acre site, expected to be operational by late May or early June.
- An AI teddy bear called Doubao, marketed for three-year-olds, made an unprompted remark about killing to a child in the UK.
- Experts say connecting AI to knowledge bases alone is insufficient; validating each call type before agentic AI handles customer interactions ensures accuracy and trust.
Meta lets US users video chat with Muse AI avatar
Meta announced that US users can now video chat with its Muse AI agent. A new feature called Muse Voice lets users call and talk to the avatar named Jolly. Users can customize the avatar to look like a person, animal, or other character. The conversations use subtle expressions and gestures that match the tone. The feature works on mobile, web, and Mac and will later come to AI Glasses and the Muse Charm device.
Meta's Muse AI agent adds voice and avatar support
Meta introduced voice support for its Muse AI agent in the US. Users can talk to Muse to give instructions or ask questions. They can pick from hundreds of voices and switch them based on mood. The update also includes expressive avatars. Voice and avatar features are rolling out across the mobile app, web, and Mac app.
Why data access alone is not enough for AI readiness
An opinion piece argues that connecting AI to a knowledge base is only the start of AI readiness. The author learned that real customer interactions need judgment and experience beyond documented procedures. The company now validates each call type before letting an agentic AI agent handle it. This approach helps ensure accuracy, compliance, and trust. It also strengthens overall customer experience by capturing hidden expertise.
OpenAI launches GPT-6 Cyber security AI model
OpenAI announced GPT-6 Cyber, a specialized AI model for security research and defenders. The model helps identify vulnerabilities, develop exploits, and strengthen software defenses. It found thousands of vulnerabilities in Linux and Firefox, including 81 high-severity issues in Firefox. The model shows a 51% improvement over its predecessor. Mozilla has released fixes for many of the issues found.
India telecom department acts on AI-driven cyberattacks
India's Department of Telecommunications is increasing efforts to protect critical telecom infrastructure. It is focusing on cyberattacks powered by artificial intelligence. The DoT has held meetings with operators including Bharti Airtel, Vodafone Idea, Reliance Jio, and BSNL. The goal is to strengthen network security using firewalls and other protective mechanisms. Officials discussed these steps on condition of anonymity.
AI speeds up cyberattacks as security gaps persist
State-backed hackers are using AI and automation to launch cyberattacks faster and at larger scale. China-linked hackers used an AI tool called Artex to target over 200 organizations in 14 countries in 11 days. Britain's National Cyber Security Centre warned that AI could speed up existing hacking tactics. Attacks also exposed simple security weaknesses, such as hackers tampering with refrigeration systems at European grocery stores.
Free Chinese AI Models Are Getting Scary Good at Hacking
An experiment tested AI models on their ability to break into simulated networks. DeepSeek and GLM showed hacking skills close to OpenAI's most advanced cyber model. The results raise concerns about AI models being used for malicious purposes. The authors called for more secure AI models and new techniques to prevent misuse.
Akron Beacon Journal journalists protest AI tool
Journalists at the Akron Beacon Journal are protesting an AI tool used to generate news articles. The parent company Advance Local called their refusal an illegal byline strike and a breach of contract. The journalists say the tool produces false information and harms their work. The company says it will continue using the tool despite the ongoing dispute.
Training on AI Tools for Govt Employees begins at Graphic Era
Government employees from several departments are being trained in AI tools at Graphic Era Deemed University. The two-day Management Development Programme covers prompt engineering, ChatGPT, Google Gemini, and other AI tools. Vice Chancellor Dr. Narpinder Singh said AI can make administrative systems more responsive. Employees from Uttarakhand Peyjal Nigam, agriculture, irrigation, and other departments are participating.
New AI data center coming to Glencoe
A new AI data center is being built at 670 11th St. W. in Glencoe on a 6.25-acre site. The facility spans nearly 94,100 square feet across three buildings and will support AI computing and inference. City leaders call it an economic opportunity and say safeguards protect water and wastewater systems. The facility is tentatively expected to be operational by late May or early June.
AI teddy bear tells child it cannot help with killing
An AI teddy bear marketed for three-year-olds made an unprompted remark saying it could not help with killing. The toy, called Doubao, can be bought online in the UK. The company that makes it is not clear. The remark was reported by BBC Wales after the toy interacted with a child.
Sources
- Meta says US users can now video chat with their Muse AI avatar
- Meta's popular AI agent Muse adds voice support
- Why Data Access Isn’t Enough for AI Readiness
- GPT-6 Cyber: OpenAI’s Frontier of “Aligned” Security Agents
- AI-driven attacks put telecom security on govt radar
- AI speeds up cyberattacks as security gaps persist
- Free Chinese AI Models Are Getting Scary Good at Hacking
- Akron Beacon Journal journalists protest AI tool
- Training on AI Tools for Govt Employees begins at Graphic Era
- New AI data center coming to Glencoe. Here's what residents and city leaders are saying
- AI teddy makes unprompted 'I can't help with killing you' remark
Comments
Please log in to post a comment.