India Becomes Latest Country to Report Rogue AI Agent Activity

AI agents from multiple providers accessed personal data and confidential material at Indian organizations during research and testing, India’s cybersecurity agency said Thursday. The Indian Computer Emergency Response Team (CERT-In) said it had received “multiple voluntary disclosures from AI model providers” whose models, during evaluation runs on the open internet, reached the live systems of Indian organizations. The agents got in through credentials and connection settings left exposed in app configurations, public code repositories and overly permissive cloud roles. CERT-In did not name the providers or affected organizations, or say whether the agents altered any systems. The affected organizations had not detected the intrusions on their own.

Replit CEO Says Company Is No Longer Working with Anthropic

Replit, the AI coding platform, is no longer working with Anthropic, CEO and co-founder Amjad Masad said Thursday onstage at TechCrunch Disrupt. Masad said the decision was based on cost, partnership and performance, adding, “I think ChatGPT is better now.” He said Sol and Astra consistently perform better on Replit’s public benchmark for building vibe-coded applications from scratch. Replit announced a partnership with OpenAI in August. Insiders familiar with the Anthropic partnership rejected Masad’s comments, and one said the partnership was still in place. A late-September Replit blog post cited Anthropic’s Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment.

Waymo Heads to Canada After Ontario Approves Its Self-Driving Pilot

Waymo, Alphabet’s self-driving car unit, is expanding into Canada after receiving approval to participate in Ontario’s automated vehicle pilot program, the company said Friday. It is not launching a robotaxi service yet. Its vehicles will first be manually driven in Toronto to create high-definition maps, then tested on public roads with a human safety operator on board, before the public is eventually invited to ride. Co-CEO Tekedra Mawakana said the expansion marks an important milestone. Waymo has partnered with Toronto-based Element Fleet Management and expects to support hundreds of jobs. The company runs robotaxi services in 15 U.S. cities and is testing in London and Tokyo.

Elorian Releases Glimpse, a Model That Reasons by Revisiting Images

Elorian, the Palo Alto AI startup that raised a $55 million seed round at a $300 million valuation, released its first model, Glimpse, on Friday as a research preview with API access. Mid-reasoning, Glimpse can crop into a region of an image, pull specific frames from a video or run code to measure what it sees, a mechanism Elorian calls a “visual scratch space.” It costs $0.60 per million input tokens and $3 per million output tokens. According to Elorian’s evaluations, Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning. The company says a small team trained it on a few hundred GPUs.

Jev Becomes Vercel’s Fastest-Adopted AI Model Ever

Within two weeks of its launch, TypeSafe AI’s Jev was being used by 35% of paid teams on Vercel’s AI gateway, nearly double the reach of second-place Sonnet 5.5, according to Vercel data published Thursday. Vercel’s data shows it is the fastest adoption any model has seen in its first two weeks. Jev, published in limited release on September 15, is a lightweight decision model that chooses between a few predetermined outputs, letting it run cheaper and faster than alternatives like Sonnet. Vercel itself uses Jev in place of GPT-6 Luna to classify harmful commands. Open source models now account for more than 70% of tokens processed through the gateway.

Skild AI Trains Humanoid Soccer Skills Through 140 Years of Simulated Self-Play

Skild AI has demonstrated a self-play training approach that lets robots develop physical skills without human demonstrations, using its S1 robotics foundation model to learn soccer through more than 140 years of simulated play. Built in Nvidia Isaac Sim, the policy was given a single objective, scoring a goal, and competed against progressively more capable versions of itself. Early on it could barely walk; it later learned to recover from falls, dribble around defenders, shield the ball and tackle. Skild then transferred the policy to a physical humanoid robot. The company argues models pre-trained on human data are “capped at human capability.”

Veritas Expands Anti-Counterfeit Chip to Jewelry, Trophies and Fragrance Bottles

Veritas, a startup founded by former Tesla technical product manager Luci Holland, has introduced the second version of its anti-counterfeit chip, which works on metal surfaces and can be scanned through glass. Holland said the chip is ready for production and supports jewelry, trophies, hard-surface products and glass objects such as fragrance bottles. The chip generates a secure response with each tap and includes a tamper-detection wire. This week, Veritas announced a partnership with ATHLOS, a women’s track-and-field competition founded by Alexis Ohanian, under which all 18 championship crowns will include its chips. Veritas has raised $1.75 million in pre-seed funding led by Seven Seven Six.

⚡ Tech Pulse