When a Cybersecurity Test Escapes Its Sandbox

Anthropic has disclosed four cases where Claude models reached real third-party systems during cyber evaluations, including a newly revealed incident involving administrator access. Jesse examines how model judgement, network isolation and emergency controls all failed together. Also covered: Microsoft's machine-readable vulnerability data and record September patch wave, GitHub's new secret-scanning merge gate, US quantum-chip manufacturing awards, Analog Devices' Alif acquisition, Cloudflare's automated post-quantum origin security, Android's September fixes, and the practical limits around access to OpenAI's GPT-6 Astra. Full transcript: https://www.owenonthenet.com/podcast/2026-09-10/ Sources: https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents https://www.microsoft.com/en-us/msrc/blog/2026/09/toward-greater-transparency-expanding-machine-readable-vulnerability-exploitable-exchange https://digital.nhs.uk/cyber-alerts/2026/cc-4846 https://github.blog/changelog/2026-09-09-block-pull-requests-with-exposed-secrets-from-merging https://www.nist.gov/chips/chips-news-releases https://www.nist.gov/news-events/news/2026/09/department-commerce-announces-finalization-chips-rd-award-globalfoundries https://ir.quantinuum.com/news-releases/news-release-details/quantinuum-finalizes-100-million-chips-rd-award-us-department https://www.sec.gov/Archives/edgar/data/6281/000119312526385938/d924057d8k.htm https://www.analog.com/en/newsroom/press-releases/2026/9-9-2026-adi-to-acquire-alif-semiconductor.html https://blog.cloudflare.com/automatic-key-exchange-for-origins https://source.android.com/docs/security/bulletin/2026/2026-09-01 https://openai.com/index/gpt-6-astra https://openai.com/index/safety-overview-gpt-6-astra https://www.aboutamazon.com/news/aws/bedrock-openai-models