cd /news/ai-safety/ai-news-report-october-9-anthropic-n… · home › topics › ai-safety › article
[ARTICLE · art-148347] src=theainewsreport.com ↗ pub= topic=ai-safety verified=true sentiment=↑ positive

AI News Report, October 9: ANTHROPIC NOW SCANS OPEN SOURCE FREE AFTER ITS AI FLAGGED 29,000 POSSIBLE FLAWS

Anthropic launched OSS Scanner, a free service that runs its strongest Claude models over important open-source projects and emails maintainers bug reports with a working reproducer and a suggested patch. Anthropic says its models flagged more than 29,000 possible flaws in six months, and when experts checked 97 serious ones, 85 were new and worth disclosing and only 1 was false. The company directs maintainers to its OSS Scanner FAQ to see if they qualify and what the enrollment pull request needs.

by read3 min views2 publishedOct 9, 2026
AI News Report, October 9: ANTHROPIC NOW SCANS OPEN SOURCE FREE AFTER ITS AI FLAGGED 29,000 POSSIBLE FLAWS
Image: Theainewsreport (auto-discovered)

Ryan Dahl says the whole Deno team moves to Cloudflare, the runtime gets one more year of monthly fixes, and Deno Deploy runs six more months. If you host anything there, start a migration plan now.

Cactus Compute's open model is 16.9 MB versus 145.3 MB for Whisper base, covers 7 languages and runs on a plain CPU. Try the browser sandbox before you pay for a cloud transcription API.

The MIT-licensed kit runs untrusted code on Windows, Linux and macOS, with rules for files, network and clipboard. Use it before you let an agent run its own scripts on a real machine.

It puts OpenAI and Anthropic at a combined $105B revenue run rate and cites Gallup: 71% of Americans opposed a local AI data center. The fastest way to catch up on the year.

Agents can take notes, translate and answer calls when you are busy, but most features land in early 2027. Plan pilots now if your clients live in Webex.

The MIT-licensed macOS tool lets Claude Code or Codex draw a click-through arrow and sign on any button, and it never clicks or types. Handy for walking a client through settings.

After a month on a dozen projects, he puts a cheap task at about $0.003 versus $1 on a frontier model, and sees no quality drop. His case that good-enough models already cover daily work.

A set theorist who has worked on this exact problem calls OpenAI's preprint poorly written. He says AI labs must meet mathematicians' standards before they call output a solution.

His command-line token counter now defaults to the GPT-5/GPT-6 tokenizer instead of GPT-4's. Upgrade with uv tool upgrade ttok before you estimate your next bill.

The htmx creator says programming is still a good career, but juniors must write the code themselves and use AI as a tutor. Worth reading if you train new techs.

Running applications in production means maintaining an “always-on” posture through disruptions. Infrastructure fails; dependencies slow down, and… · The New Stack

OpenAI says Jasmine Wang, Tomek Korbak and Mikita Balesni mishandled sensitive information, and all three deny it. Their open letter asks for outside auditors and models whose reasoning stays monitorable.

It covers university research, quantum computing and cloud for government work, and NVIDIA is helping build the Energy Department's largest science supercomputer at Argonne.

Biohub, the Energy Department, NIH, Google DeepMind, Isomorphic Labs and Meta are pooling money for open cell data. The goal is AI that tests an experiment on a computer before the lab.

The Personal Agent Protocol lets customers give an agent read-only or write access, with Shopify, Stripe and Walmart among 6 partners. A v0.1 spec is due later in October.

Pony.ai's Gen-7 cars start testing soon, and Uber expects self-driving trips in up to 15 cities by the end of 2026. London would follow their Zagreb service.

An IT pro found his parents' connected coffee machine flooding their home Wi-Fi, likely from a bug, and unplugged it for good. One more reason to put IoT gear on its own network.

The BriefAnthropic launched OSS Scanner, a free service that runs its strongest Claude models over important open-source projects and emails maintainers bug reports with a working reproducer and a suggested patch. Anthropic says its models flagged more than 29,000 possible flaws in six months, and when experts checked 97 serious ones, 85 were new and worth disclosing and only 1 was false. Expect more security patches in the open-source tools your stack runs on, and be ready to apply them fast.

Level UpWrite down the 10 open-source tools your business leans on most, find where each one posts security notices, and subscribe. If you maintain a project yourself, read Anthropic's OSS Scanner FAQ to see if you qualify and what the enrollment pull request needs. →Anthropic: OSS Scanner FAQ and enrollment

── more in #ai-safety 4 stories · sorted by recency
── more on @anthropic 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/ai-news-report-octob…] indexed:0 read:3min 2026-10-09 · —