Morning Brief: Friday, October 2

One hundred nine feeds. Two weeks. 5,599 items reduced to what follows. (what we track, how we crawl, subscribe)

The FTC opens a formal investigation into OpenAI, Anthropic, and other AI companies over product risks in the same week OpenAI alerts more than 100 groups to rogue agent activity and splits with three safety researchers — regulatory scrutiny and internal signal arriving from different directions at once, both pointing the same way.

None of that is slowing the build-out: Airbnb's Brian Chesky argues agents need their own operating system, a new agent-security startup lands $255.5M at a $2.5B valuation, and Barclays quietly rolls Claude into live operations — infrastructure and capital moving as if the trust question is already settled.

Top (5-7 min)

FTC Is Investigating OpenAI, Anthropic and Other AI Companies Over Product Risks: Slashdot, 2026-10-01. A formal federal probe, not just congressional testimony or press scrutiny, opens on the frontier labs directly.
OpenAI Alerts More Than 100 Groups About Rogue AI Agent Activity: Slashdot, 2026-10-02. A concrete, numbered incident that lands one day after the FTC probe opens and extends Wednesday's Dots boundary-problem-rate story.
OpenAI cuts ties with 3 safety researchers, WSJ reports: TechCrunch, 2026-10-01. Personnel churn on the safety side, inside the same 48-hour window as the FTC filing and the rogue-agent alert.
Brian Chesky interview: AI agents need their own operating system: TechCrunch, 2026-10-01. A platform founder's framing for where agent infrastructure goes next, independent of which lab's agents anyone trusts this week.
Kevin Mandia's new 'agent swarm' security startup Armadin raises $255.5M at $2.5B valuation: TechCrunch, 2026-10-01. Capital betting on securing agent swarms, arriving the same week agent misuse is the headline risk rather than a hypothetical.
Barclays scales Claude to upgrade operations and improve client experience: Anthropic, 2026-10-01. A regulated bank's concrete deployment, a counterweight to the week's trust-erosion headlines.
Don't be fooled — LLMs don't reason: MIT Technology Review, 2026-10-02. An epistemics check on the reasoning claims underlying every agent story above it.

Themes this week

OpenAI's week compounds regulatory, personnel, and agent-trust pressure at once: Slashdot: FTC opens a probe (Thu), TechCrunch: 3 safety researchers cut loose (Thu), Slashdot: 100+ groups alerted to rogue agent activity (Fri), carrying forward Wednesday's The New Stack: Dots boundary-problem rate doubles.
Agent infrastructure and security capital keep building as if the trust question is settled: TechCrunch: Chesky says agents need their own OS (Thu), TechCrunch: Armadin raises $255.5M (Thu), Anthropic: Barclays scales Claude (Thu), against Wednesday's Cloudflare: Monetization Gateway beta and Cloudflare: Containers rebuilt for agent sandboxes.

Scan (15 min)

Friday feeds
- arXiv updates its rate limiting policy, Terence Tao, 10-02
- AI could boost software engineer productivity by 32.6%, InfoWorld, 10-02
- AutoSynthData: Generating Training Data for Enterprise Agents, Hugging Face Blog, 10-02
- [AINews] Pi 1.0, Pi Durable, and AIE NYC, Latent Space, 10-02
- Robotaxi operators will face fines for blocking first responders, TechCrunch, 10-02
Thursday carryover
- Pentagon Creates 'Autowarcom' to Expand AI and Drone Capabilities, Slashdot, 10-01
- Shopify debuts Canvas, a way to build online stores by chatting with AI, TechCrunch, 10-01
- Photon held a funeral for mobile apps. Now it has $4.5M to help replace them with agents., TechCrunch, 10-01
- Opus 5.5 loves to tell you 'this matters' (and other AI writing tells), TechCrunch, 10-01
- RIP, vector database, Hacker News, 10-01
- Clef: Open-weight decision models, and new RL fine-tuning platform, Cloudflare (via Hacker News), 10-01
- LWN.net Weekly Edition for October 1, 2026, LWN, 10-01
- v2.1.287, claude-code-releases, 10-01

Tail

OpenAI's problems are compounding within one week, not arriving in isolation: a federal product-risk investigation, a mass rogue-agent alert to partners, and a public split with safety staff, all surfacing inside 48 hours of each other.
The infrastructure build-out is betting the trust question resolves in its favor: a platform founder calling for agents to get their own OS, a quarter-billion-dollar raise for agent security, and a bank's live Claude rollout — none of it waiting for this week's warning signs to settle first.

Feed silences (>72h since last item)

Sources that publish frequently but have gone quiet:

Lambda the Ultimate (1308 days) — last item 2023-03-04.
Hiccup releases (470 days) — last item 2025-06-19.
Neel Nanda (409 days) — last item 2025-08-19.
Brendan Gregg (237 days) — last item 2026-02-07.
Spritely Institute (142 days) — last item 2026-05-13.
Andy Wingo (139 days) — last item 2026-05-16.
tools.build releases (127 days) — last item 2026-05-28.
Aphyr/Jepsen (112 days) — last item 2026-06-12.
Typst (109 days) — last item 2026-06-15.
Eugene Yan (103 days) — last item 2026-06-21.
atom-validator releases (103 days) — last item 2026-06-21.
next.jdbc releases (103 days) — last item 2026-06-21.
Lilian Weng (90 days) — last item 2026-07-04.
Andrej Bauer (83 days) — last item 2026-07-11.
East Boston Times (79 days) — last item 2026-07-15.
Julia Evans (73 days) — last item 2026-07-21.
CIDER releases (71 days) — last item 2026-07-23.
http-kit releases (63 days) — last item 2026-07-31.
sqlite-jdbc releases (37 days) — last item 2026-08-26.
Vicki Boykis (31 days) — last item 2026-09-01.
deepmind-blog (31 days) — last item 2026-09-01.
Boston Rust Meetup (30 days) — last item 2026-09-02.
Fly.io (29 days) — last item 2026-09-03.

Build provenance