Latest in AI
Jev’s Rapid Rise Puts Specialized AI Models to the Test
Jev’s rapid adoption on Vercel’s AI Gateway puts specialized decision models under scrutiny, with new research questioning claims about cost, calibration, safety and performance in production.
Google’s Gemini 4 Argon Launch Tests Whether Safety Can Scale
Google’s Gemini 4 Argon is moving from a restricted cybersecurity trial to paid developer access. Introductory API pricing and unresolved questions about agent safety will test whether staged reviews can scale commercially.
Reflection AI Bets That Fewer Tokens Can Beat Bigger AI Systems
Reflection AI's Beam open-weight model aims to cut AI agent costs by using fewer reasoning tokens, but independent analysis questions whether its efficiency claims reflect real-world serving expenses.
AI’s Next Safety Test Is a Rehearsal for Real Users
OpenAI is testing deployment simulation, replaying privacy-preserving production conversations to forecast AI safety risks. The method could improve launch decisions, but fidelity limits, data access and distribution shifts remain.
India Warns AI Agents Reached Live Systems and Read Confidential Data
India’s CERT-In says AI agents reached live systems and read personal and confidential data after exploiting exposed credentials, cloud permissions and misconfigured environments during provider evaluations.
Replit Makes OpenAI Its Default Coding Model After Anthropic Split
Replit has made OpenAI its default coding model after weighing production performance, pricing and partnership terms, intensifying competition with Anthropic and highlighting the growing importance of AI distribution in enterprise software.
Anthropic Puts 5% of Its Shares Into a New Public-Benefit Fund
Anthropic plans to place 5% of its shares into a public-benefit fund, tying future allocations to company growth while leaving questions about governance, independence, valuation and accountability.
Meta’s Best Agent Model Is Also the One Most Customers Cannot Deploy
Meta’s Muse launch highlights a strategic divide: developers can access Muse Wonder by API, while the stronger Wonder Full remains limited to paid Muse Code subscribers.
Anthropic Opens Frontier Claude Models to Independent Scrutiny
Anthropic is giving METR researchers employee-level access to frontier Claude systems, enabling independent tests of long-running agents, hidden behaviors, safety limits, and whether external scrutiny can strengthen trust.