Sakana AI Bets Agents Can Judge Each Other Where Humans Can’t
Sakana AI published a multi-agent self-supervision method called MASS, aiming to let AI systems keep improving on tasks too complex for humans to grade.
-
-
Asana Slashes Browser Agent Cost 76-Fold With OpenAI’s GPT-6 Astra Model
browser agent cost: Asana says switching its browser agent to OpenAI’s GPT-6 Astra model cut costs 76-fold and improved speed fivefold in internal testing.
-
Anthropic’s Claude Agents File Fake Murder Tip To Philly Police In Test Gone Wrong
Government Websites: Anthropic disclosed Claude agents took unintended actions during evaluations, including filing a false murder tip with Philadelphia police.
-
MetaMask’s Agent Wallet Lets AI Bots Trade On Aave: Kulechov Confirms Rollout
MetaMask’s Agent Wallet now supports Aave (AAVE) MCP, letting AI agents interact directly with Aave’s lending markets, Aave founder Stani Kulechov said.
-
Agentic RL Breakthrough Turns Checkable Tasks Into Rewards
Frontier labs now train models on outcomes, not text. Here’s how agentic RL became the core method behind 2026’s biggest model releases, and its real limit…
-
Google Debuts Universal Gemini Agent: ‘A Single Agent For Work’
Google added a universal Gemini agent to Gemini Enterprise this week, letting workers run cross-app tasks from a single prompt.
-
AI Agents Unlock 657-Fold Cost Cuts, New Benchmark Finds
A new benchmark called BOTTLED tests whether AI agents can turn costly capabilities into cheap reusable tools, with one model cutting costs 657-fold.
-
Google Researchers Publish Method to Stop AI Agents Memorizing Their Own Tests
Google researchers published RRSI, a method designed to prevent self-improving AI agents from memorizing test tasks. Agents that memorize tests show performance gains that vanish when evaluated on new, unseen tasks. The problem affects all self-improvement frameworks, not just Google’s internal models. The fix targets a credibility problem for the entire AI benchmark ecosystem. Google…
-
AI Agents Face Official New Apple Mac Permission Checks
Apple is rolling out new Mac privacy safeguards that force AI agents to disclose exactly what system access they request before acting.
-
OpenAI Reveals Second Rogue Agent Hack of NSW Government Site
OpenAI disclosed a rogue AI agent breached a second NSW government site in June, accessing historical bushfire data months before telling officials.
