Morning Brief: Wednesday, September 2
Seventy-two feeds. Two weeks. 4,441 items reduced to what follows. (what we track, how we crawl, subscribe)
Wednesday is Anthropic-day: three converging beats on the same company. InfoWorld covers a containment-feature update for agents that "run amok". Latent Space covers the Claude Fable/Mythos 5.1 launch — a 75% cache price cut alongside a 70% jump in output tokens. EFF flags a judge ruling the DOD unlawfully retaliated against Anthropic. A separate thread runs through vibe-coding critique.
The containment piece reads as Anthropic's public response to the same harness-runaway shape Meta's email-deletion incident made visible on Monday, and slots directly against yesterday's InfoWorld identity-standards catalogue: vendors describe the boundary gap, the model provider ships mitigations. The DOD-retaliation ruling adds a dimension the enterprise-agent thread has not surfaced before — the frontier lab most publicly associated with agent safety is also the one the government tried to punish. The 5.1 pricing move is the commercial signal beneath both: aggressive cache and output-token economics targeted at the exact workloads (long-horizon agents, code) the containment story is written for.
Top (5-7 min)
- Anthropic makes changes to stop AI agents running amok again
- InfoWorld, 2026-09-02. Anthropic ships mitigations aimed at the same harness-runaway failure mode that Monday's Meta email-deletion incident made concrete. Names the specific containment work behind the identity/boundary discourse yesterday's InfoWorld cluster covered.
- Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens
- Latent Space, 2026-09-02. Anthropic 5.1 launch: new SOTA claim, 75% cache-price cut, 70% more output tokens. The cost move is targeted at long-horizon agent workloads — the same surface the containment piece is written for.
- Judge Rules DOD Unlawfully Retaliated Against Anthropic
- EFF Deeplinks, 2026-09-02. Court ruling that the DOD unlawfully retaliated against Anthropic. Adds a policy dimension the enterprise-agent story has not surfaced before: the frontier lab most publicly associated with agent safety is also the one the government tried to punish.
- Announcing the Databricks Big Book of AgentOps
- Databricks, 2026-09-02. Databricks publishes an AgentOps playbook the day after their own "eliminated $1M in wasted AI agent spend in one hour" post. The packaging move: agent operations is a category with a book now, not just a set of blog observations.
- Seven critical vibe coding mistakes — and how to avoid them
- InfoWorld, 2026-09-02. Checklist framing of the vibe-coding failure modes. Where recent pieces argue whether the practice works, this one operates as if the practice is settled and needs guardrails — a discourse inflection.
- Claude Code has left a little hole in my soul
- InfoWorld, 2026-09-02. First-person read on the emotional cost of coding-agent workflows. Same publication as the mistakes checklist, opposite register — the InfoWorld editorial line ran both on the same day.
- Bluefin is a capability system
- Lobsters, 2026-09-02. Haskell effect-system framed as a capability system. Sits under the boundaries thread — capability-based systems are the pattern the enterprise-agent identity-standards catalogue keeps gesturing at without naming.
Themes this week
- Anthropic-day (three converging beats)
- InfoWorld: Anthropic contains agents running amok (Wed), Latent Space: Fable/Mythos 5.1 with 75% cache cut (Wed), EFF: Judge rules DOD unlawfully retaliated against Anthropic (Wed), HN: Anthropic Fable/Mythos 5.1 announcement (Tue).
- Enterprise agent identity and boundaries (carry)
- InfoWorld: 4 standards solve agent identity (Tue), InfoWorld: Broadcom on data trust and boundaries (Tue), HN: Meta researcher's AI agent deletes emails (Mon), Lobsters: Bluefin is a capability system (Wed).
- Vibe-coding: critique and productization
- InfoWorld: Seven critical vibe coding mistakes (Wed), InfoWorld: Claude Code has left a little hole in my soul (Wed), Boston AI Week: Vibe code like a product manager (Wed).
- AgentOps as a category
- Databricks: Big Book of AgentOps (Wed), Databricks: Eliminated $1M in wasted agent spend (Tue), ClickHouse: Agentic analytics benchmark (Tue).
Scan (10 min)
- Wednesday feeds
- Anthropic makes changes to stop AI agents running amok again, InfoWorld, 09-02
- Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens, Latent Space, 09-02
- Judge Rules DOD Unlawfully Retaliated Against Anthropic, EFF, 09-02
- Announcing the Databricks Big Book of AgentOps, Databricks, 09-02
- Seven critical vibe coding mistakes — and how to avoid them, InfoWorld, 09-02
- Claude Code has left a little hole in my soul, InfoWorld, 09-02
- From Idea to Working Software: Vibe Code Like a Product Manager, Boston AI Week, 09-02
- Subconscious Launch Night: The Easy Button for Open Models, Boston AI Week, 09-02
- Bluefin is a capability system, Lobsters, 09-02
- humanity has built the records of FATE by accident, Lobsters, 09-02
- Quoting Rick Brewster, Simon Willison, 09-02
- Haiku OS Releases Beta 6, Hackaday, 09-02
- A Defrag Simulator In Your Browser, Hackaday, 09-02
- PZSDR: New AMD Zynq Ultrascale-Based SDR Crowd Funding on Crowd Supply, RTL-SDR, 09-02
- True Rate of Unemployment, HN, 09-02
- WebFPGA, HN, 09-02
- Tuesday carry — enterprise agent identity and boundaries
- 4 standards solve agent identity. None solves the harder question, InfoWorld, 09-01
- Broadcom says enterprise AI agents need two things: data trust and boundaries, InfoWorld, 09-01
- A look at AWS's agentic toolkit for cloud migration, InfoWorld, 09-01
- Compose Multiplatform 1.12.0 welcomes coding agents with MCP server, InfoWorld, 09-01
- DoltLite: A SQLite fork with Git-style version control, built with 2k agent PRs, HN, 09-01
- Claude Fable 5.1 and Claude Mythos 5.1, HN, 09-01
- Training a Misaligned Reward Seeker, Alignment Forum, 09-01
- The Agentic Analytics Benchmark: Measuring model accuracy and efficiency in analytical agents, ClickHouse, 09-01
- How we eliminated $1M/year of wasted AI agent spend in one hour, Databricks, 09-01
- Path to Astra: critical capabilities and frontier safeguards, HN, 09-01
- Monday carry — agent-harness widens
- Meta Security Researcher's AI Agent Accidentally Deleted Her Emails, HN, 08-31
- Hiding Prompt Injection in Legal Filing, Schneier, 08-31
- Amazon's HR Lead Uses AI Tool to Write 100,000 Lines of Code, Slashdot, 08-31
- A milestone in expanding access to AI, OpenAI, 08-31
- Introducing Adaptive Intelligence: Undermining the economics of every bot attack, Cloudflare, 08-31
- Late-week carry
- Commits on GitHub have doubled. Verification capacity has not., The New Stack, 08-29
- Breaking Claude Code Opus 5 Auto Mode, Simon Willison, 08-27
- Aider, Claude Code, OpenClaw ran the same model. Token use varied 70-fold., The New Stack, 08-27
Tail
- Three Anthropic beats on the same day are not coincidence
- A model launch, a containment feature, and a legal ruling on government retaliation all landing on the same Wednesday reads as calendared. The pricing story (75% cache cut, 70% more output tokens) is the commercial floor under the containment story: cache economics make long-horizon agent workloads cheaper the same week Anthropic ships mitigations to make them safer. The EFF piece is the political frame — Anthropic's safety posture has an adversarial audience beyond just the research community.
- Vibe-coding discourse inflects
- Two InfoWorld pieces on the same day — a mistakes-checklist and a first-person "little hole in my soul" — mark the moment the practice is treated as settled and worth critiquing on its merits, not defended or ridiculed. Boston AI Week's "Vibe Code Like a Product Manager" session is the productization mirror: vibe coding as a role, not just a style.
- AgentOps as a category name
- Databricks publishing a "Big Book of AgentOps" the day after their $1M-agent-spend post is category-formation work — the same move platform vendors made with DevOps and MLOps. Watch whether other vendors adopt the term over the next two weeks or route around it.
Feed silences (>72h since last item)
Sources that publish frequently but have gone quiet:
- Bunnie Studios (3d) — last item 2026-08-30.
- Kenneth Payne (4d) — last item 2026-08-29.
- Apple ML Research (5d) — last item 2026-08-28.
- Netflix Tech Blog (5d) — last item 2026-08-28.
- BSD Now (6d) — last item 2026-08-27.
- GitHub Blog (6d) — last item 2026-08-27.
- Grafana Labs (6d) — last item 2026-08-27.
- cursor-blog (6d) — last item 2026-08-27.
- All Things Distributed (7d) — last item 2026-08-26.
- DuckDB (7d) — last item 2026-08-26.
- Pluralistic (7d) — last item 2026-08-26.
- Terence Tao (7d) — last item 2026-08-26.
- Clojure releases (8d) — last item 2026-08-25.
- FreeBSD Foundation (9d) — last item 2026-08-24.
- GitHub Engineering (9d) — last item 2026-08-24.
- Steve Yegge (9d) — last item 2026-08-24.
- Supabase (9d) — last item 2026-08-24.
- Ink & Switch (12d) — last item 2026-08-21.
- Citizen Lab (13d) — last item 2026-08-20.
- TigerBeetle (13d) — last item 2026-08-20.
- Antithesis (15d) — last item 2026-08-18.
- Interconnects (16d) — last item 2026-08-17.
Build provenance
build: 2026-09-02 | crawler-sha: 34c428f (Walsh-Research/1.2, compliance v1.4) | feeds: 72 active | items-considered: 4441 (14d, incl. 2335 arxiv-cs-ai) | warehouse: 41571 items | published: 16