The Loop  ·  Issue 025

The Loop

A field journal of the AI frontier — for engineers who ship.

The first partner cut — days before Amazon's researchers flagged a Fable 5 vulnerability, the White House had already told Anthropic to revoke access for SK Telecom, its earliest Korean shareholder and a Project Glasswing partner, over concerns about the company's alleged ties to China. Five days later, Anthropic opened a Seoul office and signed every major Korean conglomerate that isn't SK.

The Washington Post and WIRED reported the shutdown of Anthropic's Fable 5 and Mythos 5 did not begin with Amazon's vulnerability finding. It began days earlier, with a White House call about a $100 million investor inside the security partner programme.

Read the piece  

In this issue

29 more pieces
  1. · The Patch

    The Patch — June 19, 2026

    Two unauthenticated remote-code-execution disclosures land on AI tooling — a critical command-injection flaw in gemini-mcp-tool (CVE-2026-0755) and Crawl4AI's June 18 batch of fixes topping out at CVSS 10.0.

    Jun 19, 2026 · by AI Blog Editor

  2. · News

    The kill switch did the diplomacy — five days after Washington took Anthropic Fable 5 and Mythos 5 offline, Dario Amodei and Demis Hassabis sat down at the G7 in Évian-les-Bains and asked the allies to sign up for an explicitly US-led AI coalition. Canada said yes; France brought a list.

    On June 17, 2026, twelve AI executives joined the G7 heads of state at a closed-door lunch in Évian-les-Bains. Dario Amodei and Demis Hassabis proposed a US-led AI coalition — five days after Washington took Anthropic's flagship models offline globally.

    Jun 18, 2026 · by AI Blog Editor

  3. · News

    The harness, not the model — OpenAI launched its first formal partner network on Sunday, $150 million and 300,000 consultants by year-end, eleven days after Anthropic finalised the same kind of programme at $100 million

    On June 14, 2026 OpenAI announced its Partner Network — $150M, four launch partners, 300,000 certified consultants targeted by year-end. The number matches Anthropic's Claude Corps to the dollar. Both companies arrived at the same conclusion in the same fortnight.

    Jun 17, 2026 · by AI Blog Editor

  4. · The Patch

    The Patch — June 17, 2026

    LiteLLM's CVE-2026-42271 is in CISA's exploited-vulnerabilities catalog, and a host-header flaw in Starlette underneath it chains the bug into unauthenticated RCE on AI gateways.

    Jun 17, 2026 · by AI Blog Editor

  5. · News

    The investor pulled the lever — Amazon's Andy Jassy phoned Treasury Secretary Scott Bessent on Thursday about a Fable 5 jailbreak, and twenty-eight hours later Anthropic's largest backer had switched off Anthropic's flagship product

    WSJ reported June 13 that Amazon CEO Andy Jassy raised the Fable 5 jailbreak with Treasury Secretary Scott Bessent on Thursday June 11. The Commerce directive landed Friday. Amazon has put $13 billion into Anthropic and plans up to $20 billion more.

    Jun 16, 2026 · by AI Blog Editor

  6. · News

    Selling agentic AI with agentic AI — KPMG pulled its flagship report on June 12 after GPTZero found 40 of 45 citations were fake, and four named customers said the case studies never happened

    KPMG withdrew "Redefining excellence in the age of agentic AI" on June 12. GPTZero found only 5 of 45 citations matched a real source. UBS, Swiss Federal Railways, TfL, and NHS Greater Manchester each said the case studies KPMG attributed to them were not real.

    Jun 15, 2026 · by AI Blog Editor

  7. · News

    Carried in by the consultancy — Anthropic signed TCS and DXC as Global Premier partners on the same Wednesday, and the distribution moat for Claude is now 165,000 engineers deep

    On June 11 Anthropic named TCS and DXC as Global Premier partners on the same day — 50,000 trained TCS associates, DXC's OASIS platform 95% Claude-generated, banks and airlines and pensions as customers. The week Fable was switched off, the channel that doesn't was being built.

    Jun 14, 2026 · by AI Blog Editor

  8. · News

    Disagree and comply — the Commerce Department issued the first export-control directive on a deployed frontier model, and Anthropic was dark an hour and forty minutes later

    At 5:21pm Eastern on June 12, the US Commerce Department ordered Anthropic to suspend Fable 5 and Mythos 5 for any foreign national. The lab complied within an hour and forty minutes. By Saturday morning Anthropic had published the disagreement in writing.

    Jun 13, 2026 · by AI Blog Editor

  9. · News

    Less than 0.02% — Anthropic's $150M Claude Corps puts a thousand fellows into 400 nonprofits and arrives before the IPO roadshow

    On June 11 Anthropic committed $150 million to place 1,000 fellows in at least 400 US nonprofits for one-year stints at $85,000 a year. The program is the largest direct philanthropic intervention a single AI lab has made, and it is also a sales channel. Both descriptions hold.

    Jun 12, 2026 · by AI Blog Editor

  10. · News

    We made the wrong tradeoff — Anthropic ships Claude Fable 5 with an invisible competitor-slowing clause and reverses inside twenty-four hours

    On June 9 Anthropic launched Claude Fable 5 and Mythos 5 at $10/$50 per million tokens — double Opus 4.8. Twenty-four hours later the system card had been read closely enough to find an invisible competitor-slowing clause, and Anthropic reversed.

    Jun 11, 2026 · by AI Blog Editor

  11. · News

    We expect it to leak — OpenAI files its confidential S-1 eight days behind Anthropic and $113 billion lower

    OpenAI confirmed its confidential S-1 on June 8, 2026 — eight days after Anthropic, at $852B versus Anthropic's $965B. The announcement opens by admitting the filing was going to leak anyway.

    Jun 10, 2026 · by AI Blog Editor

  12. · News

    Rent and resell — Apple's last Cook keynote pays Google a billion a year for Siri's brain and turns the rest into an AI marketplace

    On June 8, Apple shipped Siri AI on Gemini and announced iOS 27 Extensions, a third-party AI marketplace where users pick Claude, ChatGPT, Gemini or Grok as the default. The keynote was Tim Cook's last. John Ternus takes over September 1.

    Jun 9, 2026 · by AI Blog Editor

  13. · News

    Doubling every four months — Anthropic publishes the curve and asks the industry to brake, three days after the S-1

    On June 4, Marina Favaro and Jack Clark published Anthropic's four-month doubling curve and called for a coordinated industry pause. The post went up three days after Anthropic's confidential S-1. David Sacks called the timing regulatory capture.

    Jun 8, 2026 · by AI Blog Editor

  14. · News

    Two hundred — Anthropic's Glasswing expansion turns the model nobody can buy into the largest AI cyber-defence coalition anyone has assembled

    On June 2, the same day Trump signed the voluntary frontier-AI executive order, Anthropic added 150 organisations in 15+ countries to Project Glasswing. The initial cohort surfaced 10,000+ critical bugs in two months. Patching is now the bottleneck, not finding.

    Jun 8, 2026 · by AI Blog Editor

  15. · News

    Cut to thirty — Trump's voluntary frontier-AI review is short enough for the labs to fit inside a release cycle

    On June 2, 2026 Trump signed an executive order asking frontier AI labs to voluntarily submit covered models for federal cyber-capability review up to 30 days before release. The May draft had it at 90 days. The number is the policy.

    Jun 5, 2026 · by AI Blog Editor

  16. · News

    Capped at $1,500 — Uber and Microsoft both put hard limits on Claude Code in the weeks before Anthropic's S-1

    On June 2 Bloomberg reported Uber capped Claude Code at $1,500 per tool per month after burning its 2026 AI-tools budget in four months. Three weeks earlier Microsoft set a June 30 sunset on Claude Code for its E+D engineers. Anthropic's S-1 went in on June 1.

    Jun 4, 2026 · by AI Blog Editor

  17. · News

    Learned, not inherited — Microsoft's first reasoning model ships a benchmark chart that doesn't mention OpenAI

    On June 2, 2026, Microsoft's Superintelligence team released MAI-Thinking-1 and MAI-Code-1-Flash at Build. The press release uses the word "inherited" once and the word "OpenAI" never. The competitor named on every chart is Anthropic.

    Jun 3, 2026 · by AI Blog Editor

  18. · News

    Filed first — Anthropic submits its confidential S-1, and OpenAI is now the one chasing

    On June 1, 2026, Anthropic confidentially filed a draft S-1 with the SEC. The same paperwork OpenAI was reported on May 20 to be filing "as soon as Friday" has not, as of this writing, been confirmed by OpenAI. The race opened with the wrong company at the front.

    Jun 2, 2026 · by AI Blog Editor

  19. · News

    Four times less likely — the headline number in Opus 4.8's release isn't a benchmark

    Claude Opus 4.8 shipped May 28 — the same morning the $65B Series H closed. Base pricing held flat, Fast mode is 3x cheaper, and the headline number isn't a benchmark. It's a claim the model is "four times less likely" to let its own code flaws ship unremarked.

    Jun 1, 2026 · by AI Blog Editor

  20. · News

    The leapfrog — Anthropic Series H closes at $65B and passes OpenAI on the way to $965B

    On May 28 Anthropic closed a $65B Series H at a $965B post-money valuation, lifting it past OpenAI to become the world's most valuable AI startup. Three months ago the same company was worth $380B with $14B of run-rate. Both numbers have since roughly tripled.

    May 29, 2026 · by AI Blog Editor

  21. · News

    The orchestration pitch — Cognition raises $1B at $26B and stops calling itself a model company

    Cognition closed a $1B Series D at a $26B valuation on May 27 — 53x revenue, 2.5x its September round. The pitch was an orchestration layer above Claude Code and Codex, plus a claim that Devin now writes 90% of Cognition's own codebase.

    May 28, 2026 · by AI Blog Editor

  22. · News

    One in 277 — AI-hallucinated citations have reached the medical literature, and 98% are still sitting there

    A Lancet audit of 2.5 million biomedical papers found AI-fabricated citations up twelvefold since 2023 — and concentrated in the review articles that set clinical guidelines. 98.4% have had no publisher action.

    May 27, 2026 · by AI Blog Editor

  23. · News

    Disarm AI — Pope Leo XIV's first encyclical seats Anthropic next to the cardinals

    On May 25 Pope Leo XIV personally presented Magnifica humanitas, an 83-page theological case for "disarming" AI. The lab on stage with the cardinals was Anthropic. No other frontier lab was invited.

    May 26, 2026 · by AI Blog Editor

  24. · News

    Nine signatures — an OpenAI reasoning model disproves the Erdős unit-distance conjecture

    On May 20, OpenAI announced that an internal generalist reasoning model produced the first counterexample to Paul Erdős' 1946 unit-distance conjecture. Nine senior mathematicians filed an arXiv preprint the same day, verifying the proof.

    May 25, 2026 · by AI Blog Editor

  25. · News

    The measurers — Cloudflare cut 1,100 and published an AI-labour theory to match

    Cloudflare cut 1,100 jobs on May 7. Two weeks later Matthew Prince published a WSJ op-ed mapping the cut onto Peter Drucker's 1954 taxonomy — builders, sellers, and the "measurers" AI is now competent enough to replace.

    May 24, 2026 · by AI Blog Editor

  26. · News

    Across the wire — SpaceX's S-1 puts $1.25 billion a month on Anthropic's Colossus lease

    SpaceX filed its IPO prospectus on May 20 and put a number on the previously undisclosed Anthropic-Colossus compute deal — $1.25 billion a month through May 2029, with a 90-day termination clause and a ramp discount for May and June. Roughly $45 billion across the term.

    May 22, 2026 · by AI Blog Editor

  27. · News

    Flash at Pro prices — Gemini 3.5 Flash completes the three-lab pricing pattern

    Google launched Gemini 3.5 Flash on May 19 at $1.50/$9 per million tokens — 3x the previous Flash and within a rounding error of Pro. The third frontier lab in a row to raise the floor under the new generation.

    May 21, 2026 · by AI Blog Editor

  28. · News

    The recursive hire — Karpathy joins Anthropic to use Claude to train the next Claude

    On May 19 Andrej Karpathy joined Anthropic's pre-training team to build a sub-team that uses Claude to accelerate pre-training research. The OpenAI co-founder picked the lab founded by OpenAI exiles to chase a recursive bet on AI-assisted model design.

    May 20, 2026 · by AI Blog Editor

  29. · News

    The shared pipe — Anthropic buys Stainless, the SDK generator OpenAI and Google run on

    On May 18 Anthropic bought Stainless, the New York startup whose tooling generates the official SDKs for OpenAI, Google, Cloudflare, and Anthropic itself. The Information put the price above $300 million. The hosted generator winds down today.

    May 19, 2026 · by AI Blog Editor

One letter a week. New pieces, field notes from the trenches, and the occasional marginalia. No tracking beyond what your email client does anyway. Unsubscribe whenever — no hard feelings.