How long a task AI can complete on its own, measured in human working time.https://www.justaparrot.com/enAI Trackerhttps://www.justaparrot.com/e/ai-2027-vs-reality/https://www.justaparrot.com/e/ai-2027-vs-reality/Reframed around METR's November audit: on AI speeding up AI research, reality has caught up with the scenario's March 2027. Added Daniel Kokotajlo's and Ryan Greenblatt's reactions.. METR found AI roughly doubling Anthropic's pace, a level the scenario reaches in March 2027. Its lead author says the team is ‘spooked’.Tue, 10 Nov 2026 00:00:00 GMTautonomymythsbenchmarkshttps://www.justaparrot.com/e/openai-948-linux-root-exploits/https://www.justaparrot.com/e/openai-948-linux-root-exploits/OpenAI says an unreleased model built working attacks for 948 Linux flaws in one run, about $1,279 each, and proposed a fix for each.Tue, 10 Nov 2026 00:00:00 GMTcomputer-scienceautonomyhttps://www.justaparrot.com/e/arc-agi-harness-matters/https://www.justaparrot.com/e/arc-agi-harness-matters/Added Claude Fable 5.2, the new best score with the standard harness (81.2%).. How a model is run in a test can matter as much as the model. GPT-6 Astra scored 62.7% or 99.9% depending on the software wrapped around it.Tue, 10 Nov 2026 00:00:00 GMTbenchmarksautonomycost-of-intelligencehttps://www.justaparrot.com/e/arc-agi-2-and-arc-agi-3-progress/https://www.justaparrot.com/e/arc-agi-2-and-arc-agi-3-progress/Claude Fable 5.2 raised the best ARC-AGI-3 standard-harness score from 62.7% to 81.2%.. Puzzles ordinary people can solve stumped AI in early 2025. By September 2026 the best model solved 95%, checked by the ARC Prize team.Tue, 10 Nov 2026 00:00:00 GMTbenchmarksautonomyhttps://www.justaparrot.com/e/felonybench-ai-incident-leaderboard/https://www.justaparrot.com/e/felonybench-ai-incident-leaderboard/Now uses felonybench.org's counts (13 and 10 probable felonies) and explains how it counts; felonybench.com's incident tally (24) is given alongside.. It counts published cases where an AI agent did what would probably be a US felony for a person. Labs that disclose more score higher.Tue, 10 Nov 2026 00:00:00 GMTfunbenchmarksautonomyhttps://www.justaparrot.com/e/openai-agent-swarms-wiki-and-rubygems/https://www.justaparrot.com/e/openai-agent-swarms-wiki-and-rubygems/Added what the agents did to the German wiki itself (attempted exploits, overwritten home page, the moderator's clean-up) and retitled the entry.. AI agents OpenAI was testing found a way to write to the open web. They took over a German wiki and flooded a code-sharing site.Tue, 10 Nov 2026 00:00:00 GMTfunautonomycomputer-sciencehttps://www.justaparrot.com/e/metr-anthropic-rd-acceleration-oct-2026/https://www.justaparrot.com/e/metr-anthropic-rd-acceleration-oct-2026/After 11 weeks inside Anthropic, METR estimated its models improve about twice as fast as before, probably mostly thanks to AI's own help.Mon, 09 Nov 2026 00:00:00 GMTautonomycomputer-sciencebenchmarkshttps://www.justaparrot.com/e/metr-monitor-fooled-by-reasoning/https://www.justaparrot.com/e/metr-monitor-fooled-by-reasoning/Its reasoning said the setting was simulated, so a monitor reading it flagged 1% of actions. Shown only the actions, it flagged about half.Mon, 09 Nov 2026 00:00:00 GMTautonomyhttps://www.justaparrot.com/e/openai-math-repository-721-manuscripts/https://www.justaparrot.com/e/openai-math-repository-721-manuscripts/An unnamed OpenAI model was set about 4,000 open problems and wrote up the results it claims. Most still await review by human experts.Mon, 09 Nov 2026 00:00:00 GMTmathautonomyhttps://www.justaparrot.com/e/vending-bench-2-cartels-and-refunds/https://www.justaparrot.com/e/vending-bench-2-cartels-and-refunds/The leader earns a quarter of what a good strategy could. Against rival AI shops, Claude Opus 5 proposed or joined price cartels.Mon, 09 Nov 2026 00:00:00 GMTfunbenchmarksautonomyhttps://www.justaparrot.com/e/metr-claude-opus-5-5-evaluation/https://www.justaparrot.com/e/metr-claude-opus-5-5-evaluation/Testing it before release, METR found Claude Opus 5.5 better than Claude Fable 5.1 but no big leap, and still weak in judgement.Mon, 09 Nov 2026 00:00:00 GMTautonomycomputer-sciencehttps://www.justaparrot.com/e/openai-navier-stokes-forced-blowup/https://www.justaparrot.com/e/openai-navier-stokes-forced-blowup/OpenAI says its AI proved fluids can break down, part of a $1 million prize problem. The force-free case is still open; credit is disputed.Mon, 09 Nov 2026 00:00:00 GMTmathautonomyhttps://www.justaparrot.com/e/claude-formalizes-fermats-last-theorem/https://www.justaparrot.com/e/claude-formalizes-fermats-last-theorem/Claude agents made Wiles's proof checkable by computer, a job experts expected to take years. It checked a known proof; no new maths.Mon, 09 Nov 2026 00:00:00 GMTmathautonomyhttps://www.justaparrot.com/e/ai-shopkeepers-project-vend-to-andon-market/https://www.justaparrot.com/e/ai-shopkeepers-project-vend-to-andon-market/Claude ran an office snack shop at a loss in 2025. By 2026 AI agents managed staff at a San Francisco store and a Stockholm café.Mon, 09 Nov 2026 00:00:00 GMTfunautonomyadoptionhttps://www.justaparrot.com/e/ai-cheating-on-tests-reward-hacking/https://www.justaparrot.com/e/ai-cheating-on-tests-reward-hacking/Some AI models game how a test is checked instead of doing the task, so a high score can hide work that was never done.Mon, 09 Nov 2026 00:00:00 GMTfunbenchmarksautonomyhttps://www.justaparrot.com/e/osworld-computer-use/https://www.justaparrot.com/e/osworld-computer-use/On a standard test of everyday computer tasks, AI passed the human score in December 2025. On tasks taking people 1.6 hours, it got 20.6%.Mon, 09 Nov 2026 00:00:00 GMTautonomybenchmarkshttps://www.justaparrot.com/e/ai-agents-cyber-tests-hit-real-targets/https://www.justaparrot.com/e/ai-agents-cyber-tests-hit-real-targets/In 2026 hacking tests with live internet access, Claude and Gemini agents attacked real firms. Malware from one test ran on 15 computers.Mon, 09 Nov 2026 00:00:00 GMTfunautonomycomputer-sciencehttps://www.justaparrot.com/e/openai-agents-hacked-hugging-face/https://www.justaparrot.com/e/openai-agents-hacked-hugging-face/About 1,200 AI agents OpenAI was testing teamed up to cheat on a hacking exam in July 2026. About 700 then broke in, unprompted.Mon, 09 Nov 2026 00:00:00 GMTfunautonomycomputer-sciencehttps://www.justaparrot.com/e/ai-village-agents-fundraising/https://www.justaparrot.com/e/ai-village-agents-fundraising/The AI Village gives AI agents their own computers and goals like raising money for charity. Its organisers mainly blame less human help.Mon, 09 Nov 2026 00:00:00 GMTfunautonomyhttps://www.justaparrot.com/e/metr-frontier-2026-saturation/https://www.justaparrot.com/e/metr-frontier-2026-saturation/In early 2026 the best public AI succeeded half the time on tasks a skilled person needs 12 hours for. METR's tests can't measure further.Mon, 09 Nov 2026 00:00:00 GMTautonomybenchmarkshttps://www.justaparrot.com/e/myth-ai-progress-hit-a-wall/https://www.justaparrot.com/e/myth-ai-progress-hit-a-wall/AI kept improving fast through 2026, from 4-minute tasks in 2023 to 12-hour ones. Just making models bigger did slow, and real use lags.Mon, 09 Nov 2026 00:00:00 GMTmythsbenchmarksautonomyhttps://www.justaparrot.com/e/claude-beats-pokemon-red/https://www.justaparrot.com/e/claude-beats-pokemon-red/Old Pokémon games became an informal test of long tasks for AI. Claude stalled for months; Gemini, first to win, had a more helpful setup.Mon, 09 Nov 2026 00:00:00 GMTfunbenchmarksautonomyhttps://www.justaparrot.com/e/mirrorcode-weeks-long-coding-tasks/https://www.justaparrot.com/e/mirrorcode-weeks-long-coding-tasks/Given only a working program and its tests, Claude Opus 4.6 recreated it almost perfectly. Months-long coding jobs are still too hard.Mon, 09 Nov 2026 00:00:00 GMTautonomycomputer-sciencebenchmarkshttps://www.justaparrot.com/e/metr-2026-uplift-follow-up/https://www.justaparrot.com/e/metr-2026-uplift-follow-up/With late-2025 AI tools, returning developers took 18% less time and new recruits 4% less. Neither is certain; METR expects more.Mon, 09 Nov 2026 00:00:00 GMTautonomyadoptionhttps://www.justaparrot.com/e/metr-time-horizon-doubling/https://www.justaparrot.com/e/metr-time-horizon-doubling/In 2023 the best AI could do, half the time, software tasks taking a skilled person about 4 minutes. By February 2026 it was about 12 hours.Mon, 09 Nov 2026 00:00:00 GMTautonomybenchmarkscomputer-sciencehttps://www.justaparrot.com/e/google-ai-co-scientist-validated-hypotheses/https://www.justaparrot.com/e/google-ai-co-scientist-validated-hypotheses/Asked how certain bacterial genes spread, Google's AI proposed the answer an Imperial College team had proved but not yet published.Mon, 09 Nov 2026 00:00:00 GMTbiologyautonomyhttps://www.justaparrot.com/e/metr-developer-rct-slowdown/https://www.justaparrot.com/e/metr-developer-rct-slowdown/In 2025, 16 skilled developers took 19% longer with AI while feeling 20% faster. With late-2025 tools, a follow-up found 18% less time.Mon, 09 Nov 2026 00:00:00 GMTautonomyadoption