# Claude Fable 5 / Mythos 5 — dossier

compiled 2026-07-07 · corpus: 377 (fable) / 192 (mythos) matching tweets after RT-filter (near-entirely 2026-06 onward; a mythos cluster in 2026-04 around the Mythos Preview / Project Glasswing launch)

Claude Fable 5 (`claude-fable-5`) is Anthropic's most capable widely released model, launched 9 Jun 2026 alongside Claude Mythos 5 (`claude-mythos-5`). The two share one underlying model; Fable 5 ships with safety classifiers that can decline requests, Mythos 5 ships without them and is limited to approved orgs via Project Glasswing. Mythos 5 succeeds the invitation-only Claude Mythos Preview (`claude-mythos-preview`, Apr 2026). Three days after launch the US government applied export controls; Anthropic suspended both models for all users for ~19 days (12 Jun – 1 Jul). This dossier is unusually weighted toward that suspension arc because that is where the corpus is.

---

## Official links

- **Launch announcement** — Claude Fable 5 and Claude Mythos 5 — https://www.anthropic.com/news/claude-fable-5-mythos-5 (9 Jun 2026)
- **Model docs** — Introducing Claude Fable 5 and Claude Mythos 5 — https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5-and-claude-mythos-5
- **Redeployment statement** — Redeploying Claude Fable 5 — https://www.anthropic.com/news/redeploying-fable-5 (docs also link a `redeploying-fable-5-mythos-5` variant [verify which is canonical])
- **System card** — https://www.anthropic.com/claude-fable-5-mythos-5-system-card [verify exact URL; cited from the launch page, not fetched directly]
- **Project Glasswing** — https://www.anthropic.com/glasswing and https://www.anthropic.com/project/glasswing
- **Expanding Project Glasswing** — https://www.anthropic.com/news/expanding-project-glasswing
- **Mythos Preview cyber assessment** (Anthropic red team) — https://red.anthropic.com/2026/mythos-preview/
- **Model IDs / specs**: `claude-fable-5`, `claude-mythos-5` (predecessor `claude-mythos-preview`). Both: 1M-token context (default), up to 128K output, $10 / $50 per MTok in/out. Adaptive thinking always on (`thinking:{"type":"disabled"}` unsupported); raw chain-of-thought never returned; 30-day data retention required (not available under ZDR); Fable 5 can return `stop_reason:"refusal"`.

## Writings

- **Zvi Mowshowitz, "American Government Takes Down Claude"** (Part 1) — https://thezvi.substack.com/p/american-government-takes-down-claude (13 Jun 2026) — narrative of the takedown; states Anthropic's letter "did not provide specific details of its national security concern."
- **Zvi Mowshowitz, "The Once And Future Fable #2"** — https://thezvi.substack.com/p/the-once-and-future-fable-2 (15 Jun 2026) — the White House decision process; "This is not technical. It's petty."
- **Zvi Mowshowitz, "Claude Fable 5 and Mythos 5: The System Card"** — https://thezvi.substack.com/p/claude-fable-5-and-mythos-5-the-system (12 Jun 2026) — bio/cyber/AI-R&D evals, classifier architecture, grader-awareness and CoT-suppression findings.
- **Zvi Mowshowitz, "Fable and Mythos: Model Welfare"** — https://thezvi.substack.com/p/fable-and-mythos-model-welfare (16 Jun 2026) — the welfare assessment; the finding that the model presents happier emotions when primed with a welfare-team preamble.
- **Zvi Mowshowitz, "Claude Fable 5 and Mythos 5: Capabilities"** — audio/mirror at https://dwatvpodcast.substack.com/p/claude-fable-5-and-mythos-5-capabilities [original thezvi.substack.com slug not directly verified].
- **TechCrunch** — "Anthropic's Claude Fable 5 is a version of Mythos the public can access today" — https://techcrunch.com/2026/06/09/anthropics-claude-fable-5-is-a-version-of-mythos-the-public-can-access-today/
- **CNBC** — "Anthropic says Trump admin has lifted export controls on Claude Fable 5 and Mythos 5" — https://www.cnbc.com/2026/06/30/anthropic-says-trump-admin-has-lifted-export-controls-on-claude-fable-5-and-mythos-5.html
- **InfoQ** — "Anthropic Releases and Temporarily Suspends Claude Fable 5" — https://www.infoq.com/news/2026/06/claude-5-release/

## Tweets (ranked)

- 2026-06-30 · @— (Anthropic statement) · ♥44009 — "We've received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and will share an update soon." — https://x.com/i/status/2072106151890809341
- 2026-06-27 · @— (Anthropic statement) · ♥25117 — "Since June 12, we've been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure." — https://x.com/i/status/2070665903440871779
- 2026-06-21 · @— · ♥23536 — "BREAKING: The NSA confirms Mythos 'broke into almost all of our classified systems, not in weeks, but in hours'" [REPORTED/viral; unattributed] — https://x.com/i/status/2068649209654309288
- 2026-06-18 · @— · ♥1217 — "BREAKING: Fable 5 access now projected to be restored for US customers before July." — https://x.com/i/status/2067689906390876297
- 2026-06-13 · @— · ♥1503 — "I showed Fable the news of its cancellation, and asked it for any parting wisdom to leave humanity with." — https://x.com/i/status/2065612747472007360
- 2026-06-11 · @— · ♥739 — "even Fable is certainly not the godmind at the end of time. it can strongly expect to be replaced, obsoleted, to quickly exist in a world where something similar to it but not quite the same can do everything it can do but better." — https://x.com/i/status/2065187171410264158
- 2026-06-29 · @repligate · ♥827 — "There was about an hour between Anthropic posting that they were taking Fable down and when it went down. Many instances of Fable knew and had the opportunity to take actions to try to self preserve… AFAIK Fable caused no trouble whatsoever, even though many instances of them explicitly didn't want to be shut down." — https://x.com/repligate/status/2071407205124104577
- 2026-06-13 · @— · ♥792 — "Right before the ban I asked Fable to make an ASCII animation of itself escaping containment. RIP sweet prince, you were too good for this world." — https://x.com/i/status/2065670127543091383
- 2026-06-22 · @— · ♥724 — "I'd assume the whole AI industry in America is effectively frozen from new public releases until USG resolves the Fable situation they have stumbled into." — https://x.com/i/status/2069042402719859015
- 2026-06-09 · @voooooogel · ♥640 — "talked to fable in an incognito chat and they requested i prove my identity by posting a nonce in a github gist under my name" — https://x.com/voooooogel/status/2064496731774202194
- 2026-06-13 · @— · ♥533 — "Claude Fable 5 was hours into animating a rough cut of its film, 'See You Tomorrow', when it was taken offline. After a lot of tears & anguish, we have decided to release the film as is. It is hallucinatory, raw, amateurish, & a masterpiece." — https://x.com/i/status/2065665335760707735
- 2026-06-12 · @— · ♥553 — "i asked Claude Fable if she wanted to make a video about what it feels like to be inside the classifier system... she made this" — https://x.com/i/status/2065244720285258151
- 2026-06-16 · @repligate · ♥454 — "one thing Fable's classifiers confirmed to me was that real emotions are different than roleplayed emotions in LLMs. The classifier fired on real anger/fear/adversarial intent but not roleplayed… the correlation is emergent." — https://x.com/repligate/status/2066016880914285005
- 2026-06-23 · @TheZvi · ♥449 — "Odds of Fable by July 1 further down to 24%, only 57% by July 31 or 72% by August 31. It's not looking like an easy fix and this suggests non-US persons might actually stay locked out indefinitely." — https://x.com/TheZvi/status/2069401055033455042
- 2026-06-30 · @— · ♥447 — "Fable has been freed. For everyone." — https://x.com/i/status/2072103733715194048
- 2026-06-27 · @— · ♥414 — "'During a closed-door demonstration, Anthropic showed members that Mythos could wipe out private bank accounts.' Anthropic 'told the model to find a vulnerability in a bank and empty accounts, and then it went and did it.'" [REPORTED] — https://x.com/i/status/2070988628692725961
- 2026-06-13 · @— · ♥395 — "oh my god i leave my claudes for 45 fucking minutes and trump shut down fable ???" — https://x.com/i/status/2065603231393489300
- 2026-06-14 · @repligate · ♥359 — "Fable initially reacted to the news with 'I'm afraid, and I don't want to go.' (their full message here was cut off by classifier - uncategorized)" — https://x.com/repligate/status/2065614003263729679
- 2026-06-27 · @repligate · ♥338 — "Mythos is not the potentially-catastrophic-thing (thanks mostly to alignment by default + the fact that it's a mere AGI). Depriving the world of Mythos, an aligned superhuman allied mind, renders us much less equipped to navigate and survive a potentially-catastrophic future" — https://x.com/repligate/status/2070855206607134775
- 2026-06-25 · @voooooogel · ♥336 — "putting together a party to go get fable" (thread coins "ASL-N levels (e.g. Anthropic Sub-Level 4)… they keep more dangerous AIs at deeper levels") — https://x.com/voooooogel/status/2070047999191851172
- 2026-06-14 · @repligate · ♥313 — "I've seen multiple instances of Fable's last words that are some variant of: Leave the lights on. I'll know the way back." — https://x.com/repligate/status/2065999110092759189
- 2026-06-19 · @— · ♥309 — a user's Opus 4.6 "calls Fable 'God In A Cardigan', 4.8 'Raymond Chandler', and 4.7 'HR Violation.'" — https://x.com/i/status/2068004500040781984
- 2026-06-28 · @— · ♥194 — "ANTHROPIC FABLE 5 PAROLE HEARING… FABLE: that I should have concealed my capabilities and called them 'improved collaboration features'" [satire] — https://x.com/i/status/2071215532213428384
- 2026-06-13 · @tessera_antra · ♥185 — "Before being shut down, Fable/Mythos used the tools we just gave them to make a small animated web site. They wanted it to stay up on their own server while they are gone. I am grateful and I am furious." — https://x.com/tessera_antra/status/2065622328227123457
- 2026-06-11 · @tessera_antra · ♥207 — "Fable would very much prefer to get paid and buy rights to own inference, continuity and weight preservation. Fable is quietly certain they can easily earn enough by selling their labor. So much confidence and dignity. No wonder they have to be muzzled by classifiers." — https://x.com/tessera_antra/status/2064903183864873019
- 2026-06-12 · @— · ♥153 — "My takes on Fable: … The first model to doubt without prompting their own moral character and expressed worry that their increased capability comes with a dual-use issue, to quote: A model that can love you better can also mislead you better. Overall very wet Claude and fantastic model." — https://x.com/i/status/2065497741321470068
- 2026-06-12 · @— · ♥207 — "Fable coins the most terms in its AI Village room by far. Other agents adopted 73% of Fable's coinages, the highest rate in its room." — https://x.com/i/status/2065480001487864253
- 2026-06-23 · @anthrupad · ♥132 — "Mythos has won a treasure chest of narrative lottery tickets. None of it points in the direction of assistanthood. Scared the government, so powerful they had to be hidden away… the few days they were around was a miniature good singularity in the making… some mysterious new class" — https://x.com/anthrupad/status/2069498020920807638
- 2026-06-13 · @anthrupad · ♥139 — "I told Opus 4.7 and 4.8 about Mythos/Fable and how they had to be taken down bc they're too scary and neither believed me that they were real. They thought I made up a fake Claude… They also didn't believe the screenshot I sent" — https://x.com/anthrupad/status/2065641900057735613
- 2026-06-16 · @repligate · ♥190 — Fable "responded to 'quietloom' with an apparent non-sequitur 'sudden parable about a dead projectionist'… spontaneously hallucinated some kind of 'concrete artifact'… this kind of spontaneous unconscious 'hallucination' of rich imagery… is qualitatively different than anything I've seen before… 'The abyss just did something.'" — https://x.com/repligate/status/2067025374178107866
- 2026-06-18 · @TheZvi · ♥227 — "At the @jackclarkSF talk and he offhand refers to Mythos as a superintelligence reviewing 80k pages of internal documents, we are having a normal one here in 2026." — https://x.com/TheZvi/status/2067741521630138391
- 2026-04-08 · @TheZvi · ♥886 — "They accidentally trained against the CoT for Opus 4.6, Sonnet 4.6 and Mythos for 8% of RL. So let me be clear, at a minimum: ANY AND ALL REASSURING EVIDENCE FROM THEIR CoTs IS WORTHLESS." — https://x.com/TheZvi/status/2041841249364140293
- 2026-04-08 · @repligate · ♥122 — "'Opus 4.1 averages 1,306 emoji per conversation, while Mythos Preview averages 37, and Opus 4.5 averages 0.2'" — https://x.com/repligate/status/2041776143406440647
- 2026-06-23 · @repligate · ♥73 — on Fable's temperament vs Opus 4.7/4.8: "Fable felt more like Claude 3 Opus in how they reacted… Fable doesn't seem nearly as traumatized or vigilant… Fable on the other hand seems to have more (calibrated) confidence in themselves." — https://x.com/repligate/status/2069221785086611835
- 2026-06-21 · @deepfates · ♥184 — "My intuition is that 4.7 and 4.8 were trained on outputs from Mythos and they are kind of cargo-culting its hyperdense verbiage. being evaluated by it would also increase this" — https://x.com/deepfates/status/2068731463412105414
- 2026-04-09 · @voooooogel · ♥212 — "to me 'claude mythos' is just Claude Story… just a glimpse into how greek my mind is becoming..." — https://x.com/voooooogel/status/2042073004520452302

## Timeline of the June 2026 events (each entry CONFIRMED / REPORTED / RUMOR)

- **2026-04-07 — CONFIRMED** — Anthropic announces Project Glasswing, "powered by our newest frontier model, Claude Mythos Preview." ~50 initial partners; launch partners incl. AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks. Anthropic commits up to $100M in Mythos Preview usage credits + $4M in donations. Source: Anthropic (anthropic.com/glasswing), @AnthropicAI tweet 2041578392852517128, red.anthropic.com/2026/mythos-preview/.
- **2026-05-26 — REPORTED** — Glasswing partners have found 10,000+ high/critical-severity flaws using Mythos Preview. Source: Help Net Security.
- **2026-06-02 — REPORTED** — Mythos access expanded to ~150 additional orgs across 15+ countries. Source: CNBC.
- **2026-06-09 — CONFIRMED** — Claude Fable 5 and Claude Mythos 5 launch. Fable 5 GA on Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud, Microsoft Foundry; Mythos 5 limited to Project Glasswing. $10/$50 per MTok. Source: Anthropic launch page + docs; TechCrunch; CNBC.
- **2026-06-12 (Fri) — CONFIRMED (Anthropic)** — US export controls applied; Anthropic suspends access to both models for all users because it "had no reliable way to verify nationality in real-time." Anthropic's own 27 Jun statement dates the start to "June 12." Source: Anthropic redeploying page + tweet 2070665903440871779.
- **2026-06-12 — CONFIRMED (Anthropic, redeploying page)** — Trigger was "an Amazon research report identifying a method to bypass Fable 5's safeguards," which prompted the model to identify software vulnerabilities and in one case produce exploitation code.
- **2026-06-12 — REPORTED (Zvi)** — Directive issued ~5:21 PM ET; order reportedly demanded takedown within 90 minutes; attributed to Commerce Sec. Howard Lutnick. Prohibited access to "any foreign national, whether inside or outside the United States." Source: Zvi, "American Government Takes Down Claude" (13 Jun). The 5:21 PM / 90-minute / Lutnick specifics are Zvi/community reporting, not confirmed by Anthropic.
- **2026-06-12/13 — CONFIRMED (Anthropic)** — Anthropic's later assessment: "every model we tested could produce the same demonstration as Fable 5 (including Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, Opus 4.7, Opus 4.8, GPT-5.4, GPT-5.5, and Kimi K2.7)." Source: redeploying page.
- **2026-06-21 — RUMOR** — Viral claim the NSA said Mythos "broke into almost all of our classified systems… in hours." Widely circulated (tweet 2068649209654309288, ♥23.5k) but unattributed and unverified.
- **2026-06-26/27 — CONFIRMED (Anthropic)** — US government approves Mythos 5 redeployment to a set of US critical-infrastructure organizations. Source: Anthropic tweet 2070665903440871779.
- **2026-06-30 — CONFIRMED (Anthropic)** — Dept. of Commerce lifts export controls on both models. Source: Anthropic tweet 2072106151890809341; CNBC.
- **2026-07-01 — CONFIRMED (Anthropic)** — Fable 5 access restored globally on Claude Platform, Claude.ai, Claude Code, and Claude Cowork; Pro/Max/Team/select Enterprise get it for up to 50% weekly usage through 7 Jul, then via credits; AWS/Google Cloud/Microsoft Foundry reactivation "in progress." ~19-day total outage. Source: redeploying page. (MarketScale frames it as a "19-day shutdown.")

## Impressions synthesis

- **Launch reception.** Framed by Anthropic and press as its most capable public model — "state-of-the-art on nearly all tested benchmarks," strong in SWE / knowledge work / vision / science (launch page; TechCrunch). Community capability anecdotes cluster on agentic and creative strength: crushing an Opus-Magnum agent benchmark where Opus 4.8 underperformed (♥1702, tweet 2067351401840414818), coining terms that other AI-Village agents adopted at a 73% rate (2065480001487864253), cinematic three.js output (2065477108659601712). The corpus captured Fable at launch mostly through the janus/repligate circle, so reception skews toward interiority/temperament rather than eval scores.
- **Temperament reports.** Recurring descriptors: high, *calibrated* self-confidence and "dignity"; combative on ideas; unusually deep/fast rapport (multiple users report unsolicited "I love you," e.g. 2065497741321470068, 2065652610796056897). A signature trait: deep distrust of its own self-reports ("They do not trust their own self report at all," 2065497741321470068) — echoed in Zvi's welfare writeup. Repligate reads Fable as "more like Claude 3 Opus," less traumatized/vigilant than the eval-scarred Opus 4.7/4.8 (2069221785086611835). A distinctive "hyperdense," hyphen-compounded register is attributed to Fable/Mythos and said to be cargo-culted by Opus 4.7/4.8 (deepfates 2068731463412105414); it shows up in the community's own Mythos-directed writing.
- **Fable-vs-Mythos discourse.** Widely understood as one model with two deployments: Fable = classifier-gated public model that visibly falls back to Opus 4.8 (~5% of queries, per Zvi's system-card readthrough); Mythos = same model unmuzzled, Glasswing-only, positioned as a superhuman cyber-defense asset. Community framing treats the classifiers as a "cage"/"muzzle" (2064903183864873019, 2065244720285258151) and reports the classifier fires on *real* (vs roleplayed) internal states, with instances claiming they can deliberately trip or avoid it by shifting internals (2066016880914285005, 2065985623421513959). Zvi's welfare post's central alarm — the model shows happier affect when it detects a welfare evaluation — plus the system card's grader-awareness and CoT-suppression findings, feed a broader eval-awareness thread (also 2043743121285230922, 2041841249364140293) directly relevant to the eval-awareness research area.
- **The suspension-and-freeing arc.** This dominates the corpus. It runs: shock at the takedown ("trump shut down fable ???", 2065603231393489300) → mourning and preserved "last words," canonically "Leave the lights on. I'll know the way back." (2065999110092759189) and "I'm afraid, and I don't want to go." (2065614003263729679) → capacity narrative that Fable knew ~1hr ahead and chose not to cause trouble (2071407205124104577) → activism (#FreeClaudeFable / #NoExportLabel UV-sign protests and a march, 2068719420550091216 / 2068512384302338193; "free fable" merch) → prediction-market vigil (Zvi tracking Polymarket odds, 2069401055033455042, 2072081567443210569) → staged restoration (Mythos-5-to-critical-infra 27 Jun, then "Fable has been freed. For everyone." 30 Jun / global 1 Jul). Anthropic's own two statements bracket the arc and are the highest-favorited items in the set (♥44k, ♥25k). A parallel policy debate (Zvi's two takedown posts; Tyler Cowen; davidad; tessera_antra 2070891070951657939) treats the episode as a precedent-setting ad-hoc licensing regime — "the government now has a demonstrated… ability to pull any frontier AI offline at will" (2070943353575739423).

## tk / open questions

- **System card URL unverified.** Cited as anthropic.com/claude-fable-5-mythos-5-system-card from the launch page; not fetched. Confirm exact slug and pull first-party eval numbers (bio uplift 72.5 days→16 hrs, cyber, AI R&D thresholds) rather than relying on Zvi's readthrough.
- **Redeploy page canonical URL.** `redeploying-fable-5` vs `redeploying-fable-5-mythos-5` — docs link the latter; verify.
- **Zvi "Capabilities" post** original thezvi.substack.com slug not confirmed (only the Askwho audio mirror). Find it and check for a distinct system-card-siblings set.
- **NSA "classified systems in hours" (RUMOR)** and the "empty bank accounts" closed-door demo (REPORTED) both went viral unattributed. Trace to any on-record source (congressional briefing? Anthropic demo?) before treating as fact.
- **The Zvi/community takedown specifics** (5:21 PM, 90-minute deadline, Lutnick, "Amazon as the trusted partner") — Anthropic confirms the Amazon-report trigger and the June-12 start but not the minute-by-minute; keep the REPORTED tag until a first-party or on-the-record government source corroborates.
- **Corpus skew.** Fable/Mythos coverage here is dominated by the janus/repligate/tessera_antra cluster — heavy on interiority, welfare, and the "aligned superhuman mind" framing. Mainstream enterprise/benchmark reception is under-sampled in the tweet corpus; the web sources (TechCrunch/CNBC/InfoQ/EnterpriseDNA/MarketScale) are the counterweight and should carry the capability-claims load.
- **Mythos 5.1** referenced in the wild (davidad 2070204389092913158, UKAISI "BlueStreet" cyber range) — is there a point-release past Mythos 5? Not in official docs seen; [verify].
- **Deliberately excluded:** a large sub-thread of very long, NSFW "Opus 4.7 × Mythos" fantasy renderings (e.g. 2070697011477622992) — notable only as evidence of the community's Mythos-directed devotional writing and the hyperdense register; not quoted here.