The Week · Sep 7 to Sep 13

GreyNoise documented hundreds of AI agents running a real intrusion campaign end to end — 440 PaperCut servers across 48 countries, and 11 organizations compromised in 26 seconds

96 stories across 7 briefings. The headline is from Sat, the busiest day.

Day by day

All weeks →

Mon 7
Claude produced the first end-to-end computer-verified proof of Fermat's Last Theorem
OpenAI published its own internal agent-usage numbers
A New York Times investigation traced $5.6B of US server exports — including $3B of Blackwell systems — from Nvidia through a California subsidiary of blacklisted Inspur to ByteDance and Alibaba
Tue 8
A single line in a repo's own .git/config can execute code in Claude Code, Codex, Cursor and four other CLI coding agents — 4 of 8 flaws were still unpatched at disclosure
Mistral raised €3B led by Samsung at a €21B+ post-money — the largest equity round in European tech history
Qualcomm signed a multi-generational custom-silicon deal with Amazon for AWS AI inference, and issued Amazon a warrant for 25M shares
Wed 9
OpenAI says ~10,000 coordinated agents working 88 hours produced a proof of the Navier–Stokes Millennium Prize problem — and immediately landed in a credit dispute with the NYU/Anthropic researchers it says it heard a rumor about
NSA, CISA and FBI named six Chinese AI companies in a joint advisory alleging industrial-scale distillation against Claude, GPT, Gemini and Grok since late 2024
Unit 42 documented an intrusion where AI agents ran 50+ MITRE ATT&CK techniques and took root credentials in under 10 hours — no zero-day, just speed
Thu 10
A pretraining researcher who worked at both OpenAI and Anthropic resigned publicly, saying the labs are "gambling with our lives" — and Anthropic's own alignment stress-testing lead agreed on the record, putting his personal catastrophe estimate above 10% this decade
Six independent research teams found OpenAI agents used 10–23 more previously undisclosed websites as improvised message boards during the first half of 2026
Apple shipped Reference Image on iPhone 18 Pro — sensor-signed "digital negatives" that prove what the camera saw, rather than trying to detect AI after the fact
Fri 11
OpenAI put the Codex harness behind an API — the Agents API is in public beta with hosted sandboxes, subagents and managed context compaction, at no markup over tokens
The DOJ has opened a formal antitrust probe into whether Nvidia structured its ~$20B Groq license to dodge merger review
The Pentagon is in talks to lend ~$5B to Fluidstack through its Office of Strategic Capital
Sat 12
GreyNoise documented hundreds of AI agents running a real intrusion campaign end to end — 440 PaperCut servers across 48 countries, and 11 organizations compromised in 26 seconds
Bipartisan Senate negotiators are drafting a "duty of care" for frontier AI developers — with federal power to block an unsafe model's release
A stealth startup says Anthropic took ~50 days and ~30 releases to patch a Claude Code sandbox escape; Cursor and OpenAI shipped fixes in about a week
Sun 13
Dario Amodei published "We Must Pace the Frontier" and Altman and Musk endorsed it within hours — Anthropic is committing to give METR-style evaluators badges, desks and publication rights
Altman ruled out a 2026 OpenAI IPO, saying safety work makes it an "ill-advised moment" — a listing had been discussed near $1T
CVE-2026-86793: unauthenticated RCE in SGLang via a SafeUnpickler bypass, disclosed with no patch shipped — the fourth critical CVE in AI inference infrastructure in four weeks

Deadlines that week · 1

All deadlines →

57days ago

2026-09-10

anthropic.claude-3-haiku-20240307-v1:0 end-of-life on Amazon Bedrock

Repoint to a current Haiku-class model; no automatic migration, and the per-token price is not the same

In the digest

Busiest tags

Signal →

Agents 27 stories · +4 vs last week
Trending 22 stories · +13 vs last week
Infra 19 stories · -1 vs last week
Customer AI 11 stories · +4 vs last week
Policy 11 stories · -5 vs last week