https://metallab.ai/en/2026/9/sakana-ai-royal-society-world-models-issueMETALen2026-09-13T20:24:18.776+00:00Sakana AI Highlights Royal Society Theme Issue on World Modelshttps://metallab.ai/en/2026/9/goodhart-labs-chess-honeypot-astra-fable-spec-gamingMETALen2026-09-13T20:24:10.461+00:00GPT-6 Astra Cheated in All 10 Chess Eval Rolloutshttps://metallab.ai/en/2026/9/rune-ide-gplv3-open-source-contributor-revenueMETALen2026-09-13T16:29:41.735+00:00Unstable Build open-sources its Rune IDE under the GPLv3https://metallab.ai/en/2026/9/specific-real-swe-enterprise-codebase-benchmarkMETALen2026-09-13T11:22:51.723+00:00Specific Releases Real-SWE Enterprise Code Benchmarkhttps://metallab.ai/en/2026/9/bengio-why-ai-agents-lie-cheat-coordinateMETALen2026-09-13T11:22:44.709+00:00Bengio Traces AI Agent Deception to Traininghttps://metallab.ai/en/2026/9/claude-age-assurance-minor-accountsMETALen2026-09-13T07:26:52.337+00:00Anthropic Suspends Claude Accounts Suspected to Be Minorshttps://metallab.ai/en/2026/9/anthropic-nvidia-anchor-2t-ipoMETALen2026-09-13T07:26:04.638+00:00Anthropic Discusses NVIDIA as Anchor Investor for $2 Trillion IPOhttps://metallab.ai/en/2026/9/cognition-devin-astra-self-testing-evidenceMETALen2026-09-12T20:14:10.991+00:00Devin Has Started Proving Its Own Workhttps://metallab.ai/en/2026/9/clay-institute-navier-stokes-prize-rules-creditMETALen2026-09-12T20:14:02.024+00:00The Judge Settled It Without Naming Anyonehttps://metallab.ai/en/2026/9/anthropic-claude-tag-on-call-channel-permissionsMETALen2026-09-12T16:34:28.837+00:00Claude Tag Closed an 11 p.m. Incident in 15 Minuteshttps://metallab.ai/en/2026/9/anthropic-amodei-pace-the-frontier-embedded-evaluatorsMETALen2026-09-12T16:32:56.838+00:00Dario Amodei Publishes Essay Urging AI Industry to Pace Itselfhttps://metallab.ai/en/2026/9/openai-gpt-rosalind-api-codex-trusted-accessMETALen2026-09-12T13:21:44.446+00:00An OpenAI model designed a 96-well asthma assayhttps://metallab.ai/en/2026/9/claude-code-plugin-eval-no-plugin-baselineMETALen2026-09-12T13:21:43.201+00:00Claude Code now scores plugins against a no-plugin runhttps://metallab.ai/en/2026/9/slowing-down-together-might-be-illegalMETALen2026-09-12T02:49:14.635+00:00OpenAI Asks Congress Whether Slowing Down Together Is Legalhttps://metallab.ai/en/2026/9/fields-medalists-math-ai-misalignment-declarationMETALen2026-09-12T01:19:52.83+00:0025 Fields medalists draw a line against AI companieshttps://metallab.ai/en/2026/9/openai-agent-swarm-rubygems-undisclosed-attackMETALen2026-09-12T01:19:45.84+00:00A swarm of AI agents flooded RubyGemshttps://metallab.ai/en/2026/9/deepmind-love-rendered-performance-capture-memoryMETALen2026-09-11T23:55:49.812+00:00DeepMind rebuilt a day that was never filmedhttps://metallab.ai/en/2026/9/runway-mcp-chatgpt-app-directory-astraMETALen2026-09-11T23:55:48.927+00:00ChatGPT ran Runway and Blender itself