A quiet weekend for vendor releases. No new changelog entries from OpenAI, Anthropic, Google, Cursor or GitHub, and no Codex or ChatGPT limit changes.
Top line
GitHub Actions and webhooks were delayed for almost two hours before today's brownout
GitHub rated a "major" incident from 4:00 to 5:50 a.m. CT Monday, with delayed Actions run starts and webhook deliveries. Actions was mitigated at 5:06 a.m. and webhooks at 5:45 a.m. Earlier, from 11:59 p.m. Sunday to 3:09 a.m. CT, logged-out and unauthenticated traffic saw 504 errors, and GitHub warned that "CI jobs and release downloads may fail" if they make anonymous calls. Authenticated requests weren't affected. It's the second night in a row of errors for unauthenticated traffic. GitHub says root-cause reports will follow; the Nov. 5 Actions report is still unpublished.
Why it matters: Builds that ran overnight may have stalled or failed. Today's planned pull_request_target block on public repos runs 10 a.m. to noon CT, so separate those failures from leftover incident fallout before you debug.
Sources: Actions and webhooks incident · Anonymous-traffic incident · GitHub changelog
Correction: Claude Code's team did respond to the secrets-leak report
Our Nov. 8 edition said Anthropic had not responded publicly to Apiiro's finding that Claude Code's .claude/settings.local.json approval file had leaked secrets into public repos. That was wrong. On Nov. 7 at 3:13 p.m. CT, Boris Cherny of the Claude Code team wrote on Hacker News that Claude Code "has not done secret scrubbing" but the team is "looking into what we can do to prevent this." He said the file is automatically gitignored when Claude creates it, recommended auto mode and secret scanning, and later said the team would prototype options "before we ship something." A commenter writing as part of the research team disputed the cause, saying that in 539 of the 1,105 repos "we traced," a commit co-authored by Claude changed the file. This is an employee comment, not an official advisory, and no fix has shipped.
Why it matters: The advice doesn't change. Rotate any secret that ever appeared in these files and turn on secret scanning.
Sources: Cherny's comment · Follow-up comment · Researcher's reply
Catch-up: Microsoft says attackers use an AI agent to find leaked secrets on GitHub
In a Nov. 5 post we missed, Microsoft Threat Intelligence said attackers use an "agentic AI offensive framework" it calls ARTEX to search public GitHub repos for credentials, including AI inference API keys. In one case, an agent went from scanning GitHub to taking data out of Azure Key Vault "in just five minutes." Sign-ins began as early as Oct. 28, including a stolen multitenant vendor app credential used against downstream tenants. A Nov. 6 follow-up says a financially motivated group Microsoft tracks as Storm-3358 is using ARTEX against financial firms, mainly in Southeast Asia.
Why it matters: Microsoft's words: "removing the original disclosure alone does not remediate the exposure." Leaked keys now get found and used within minutes, so rotation is the only fix.
Sources: Microsoft Threat Intelligence, Nov. 5 · Microsoft follow-up, Nov. 6
What to do
- Check overnight CI and webhook consumers. Re-run Actions jobs that stalled between 4:00 and 5:50 a.m. CT, look for missing webhook deliveries, and make CI downloads authenticated. GitHub status
- Azure tenants: hunt for ARTEX activity. Review service-principal sign-ins against the indicators in Microsoft's post and move apps off secret-based credentials where you can. Microsoft Threat Intelligence
- Self-managed Firecracker or KVM hosts: patch the host kernel and reboot. A public exploit chain now turns CVE-2026-99999 into a guest-to-host escape; AWS says EC2, Lambda and Fargate are not affected. AWS bulletin
- KEV deadlines this week: Adobe Commerce CVE-2026-48356 and Brotli CVE-2026-111497 on Nov. 10, VMware ESX CVE-2026-41709 on Nov. 11, and Ivanti Connect Secure CVE-2026-118527 on Nov. 13. CISA KEV
Also notable
- China CAC: one filing portal for AI services. A single system for generative AI, algorithm, deep-synthesis and blockchain filings went live Nov. 9 at beian.cac.gov.cn, and the old algorithm and blockchain filing systems were shut down. CAC notice
- ByteDance: Seedream 5.5 image model. It keeps unedited areas unchanged across up to 100 rounds of editing and outputs transparent backgrounds. It's rolling out in Doubao, Jimeng and Volcano Ark, but API pricing isn't on the official post. ByteDance Seed
- Catch-up: PwnEscape Part 2. Researchers turned the kernel bug into a Firecracker escape in a demo run as root without the jailer. A separate Firecracker bug, CVE-2026-122192, is fixed in 1.16.2 and 1.17.0. PWN Research write-up
- OpenAI: ChatGPT image generation had elevated errors from 9:11 to 11:00 a.m. CT Sunday. The Nov. 6 incident with declined U.S. business Mastercards is still marked monitoring. OpenAI status
- Watch: Inkling 2 Small and Hy4 full weights are still unpublished. Inkling 2 Small
Sources
Actions and webhooks incident · Anonymous-traffic incident · GitHub changelog · Cherny's comment · Follow-up comment · Researcher's reply · Microsoft Threat Intelligence, Nov. 5 · Microsoft follow-up, Nov. 6 · AWS bulletin · CISA KEV · CAC notice · ByteDance Seed · PWN Research write-up · OpenAI status · Inkling 2 Small