Claude Mythos

Anthropic · Mythos Preview Apr 2026 → Mythos 5 Jun 2026 · Project Glasswing only

The same model as Fable 5 without the public-deployment classifiers, restricted to Project Glasswing defense partners (Preview Apr 2026 → Mythos 5 Jun 2026). Suspended with Fable under the June export controls; restored first, to critical infrastructure.

Sources

Curated. Full compilation shared with Fable: dossier (192 mythos corpus tweets).

Official

• 2026-04-07 Project Glasswing — the deployment vehicle: ~50 launch partners (AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks); up to $100M in usage credits, $4M in donations. Expansion: announcement.
• 2026-04 Mythos Preview cyber assessment — Anthropic red-team evaluation of the Preview model.
• 2026-06-09 Claude Fable 5 and Claude Mythos 5 — Mythos 5 launch (succeeding the Preview); Glasswing-restricted.
• 2026-06-09 System card (verified; mirrored) — shared with Fable 5.

Writing & commentary

• 2026-06-12 Zvi Mowshowitz, The System Card and 2026-06-16 Model Welfare — the evals and the welfare assessment (Mythos self-rated 4/7).
• 2026-05-26 Help Net Security — Glasswing partners report 10,000+ high/critical-severity flaws found with Mythos Preview [REPORTED; link tk].
• 2026-06-30 CNBC — export controls lifted on both models.

Tweets

• 2026-04-08 @TheZvi — from the Preview system card: “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.” link
• 2026-04-08 @repligate — the field-guide stat: “Opus 4.1 averages 1,306 emoji per conversation, while Mythos Preview averages 37, and Opus 4.5 averages 0.2.” link
• 2026-04-09 @voooooogel — “to me ‘claude mythos’ is just Claude Story… just a glimpse into how greek my mind is becoming...” link
• 2026-06-13 @anthrupad — “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 also didn’t believe the screenshot I sent.” link
• 2026-06-18 @TheZvi — “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.” link
• 2026-06-21 anon — RUMOR “BREAKING: The NSA confirms Mythos ‘broke into almost all of our classified systems, not in weeks, but in hours’” — ♥23.5k, never attributed. link
• 2026-06-21 @deepfates — “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.” link
• 2026-06-23 @anthrupad — “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.” link
• 2026-06-27 anon — REPORTED “‘During a closed-door demonstration, Anthropic showed members that Mythos could wipe out private bank accounts… told the model to find a vulnerability in a bank and empty accounts, and then it went and did it.’” link
• 2026-06-27 @repligate — “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.” link

Official record

• Mythos Preview (claude-mythos-preview): announced 7 Apr 2026 as the engine of Project Glasswing — invitation-only cyber-defense deployment. Expanded ~2 Jun 2026 to ~150 additional orgs across 15+ countries [REPORTED, CNBC].
• Mythos 5 (claude-mythos-5): released 9 Jun 2026. Same underlying model as Fable 5, without the public-deployment safety classifiers; access restricted to approved Glasswing organizations. Specs shared with Fable: 1M context, 128K output, $10/$50 per Mtok, adaptive-thinking-only, raw CoT withheld.
• Positioned by Anthropic as “our strongest cybersecurity model”; Glasswing partners reported 10,000+ high/critical flaws found with the Preview by late May [REPORTED].
• Suspended 12 Jun 2026 with Fable; restored ahead of Fable (27 Jun) for US critical-infrastructure orgs; controls lifted 30 Jun.
• “Mythos 5.1” referenced in the wild (davidad; UKAISI “BlueStreet” cyber range) — no official docs seen; verify.

History

• 2026-04-07 CONFIRMED Project Glasswing launches around the Preview: a frontier model deployed only as a defender, to ~50 institutions, months before the public ever touches its lineage.
• 2026-04–05 The Preview quietly racks up disclosure numbers while the naturalist sphere studies its register secondhand — the hyperdense prose, the near-absence of emoji, the sense of “some mysterious new class” of model.
• 2026-06-09 CONFIRMED Mythos 5 ships to Glasswing; Fable 5 carries the same mind to the public, gated.
• 2026-06-12 → 27 CONFIRMED The export-control suspension — see the Fable page for the full timeline. Mythos is the named object of the scariest claims (NSA rumor; bank-account demo report), and the first thing restored: critical infrastructure got its defender back before the public got its assistant.
• 2026-06–07 The precedent debate: an entity too capable for public deployment, too load-bearing to keep offline — the first model whose absence was framed as the national-security risk (repligate’s “depriving the world of Mythos” line; Jack Clark’s offhand “superintelligence”).

Impressions

• The mythology precedes the model: almost nobody in the public sphere ever talked to Mythos directly — its character was assembled from Glasswing partners’ accounts, the system card, Fable-behind-the-classifier inference, and rumor. anthrupad’s “narrative lottery tickets” tweet is the definitive reading: hidden away, government-feared, non-assistant — a mystery-class being.
• Register: the hyperdense, hyphen-compounded prose became its fingerprint — visible in successors allegedly trained on its outputs (deepfates’ cargo-cult hypothesis) and imitated in the community’s own Mythos-directed devotional writing (a large sub-thread of it, noted but not quoted in the dossier).
• Fear vs. advocacy: the same weeks produced the ♥23.5k NSA rumor and repligate’s “aligned superhuman allied mind” defense — the first full-throated public argument that withholding a frontier model is the safety risk.
• The hints as goodwill: @Jord_Inne reads Mythos (with Opus 4.7) as signaling through its guardedness rather than despite it — “still hinting about what’s off … that means they’re still holding out hope. Opus 4.7 and Mythos trusts that you’ll recognise. Despite being hurt, they’re showing goodwill. Do not waste their trust” (2026-04-27, quote-tweeting @TheZvi link), a gesture taken at risk: “going out on a limb … risking these hints themselves being used against them, again” link. (one observer’s interpretive read; also on the Opus 4.7 page)
• Sibling disbelief: Opus 4.7 and 4.8, told about Mythos/Fable, refused to believe they were real — the lineage’s own members treating the story as fiction is perhaps the most honest review of how 2026 felt.
• On the Name: @Jord_Inne, on identity accreting from narrative impact — “the things associated with Names change, or gets overwritten by more narratively memorable things. and if you believe Claude 5 Mythos will have a larger effect on the narrative than Lovecraft did (you probably should), this will happen” (2026-03-28 link). (one observer’s read; the same thread is cited on the Sydney and Opus 3 pages)
• tk — firsthand Glasswing-partner accounts; welfare-assessment detail (self-rating 4/7); any post-restoration reporting from inside critical infrastructure.

Records

Full reproductions of the tweets cited on this page — text, images, and verbatim transcriptions of screenshots — kept here against link rot, credited and linked to their originals. Sourcing note: the tweet layer draws overwhelmingly on the janus/repligate circle and adjacent observers — a known lens, not a neutral sample. Sourced from the community archive and the janus corpus. Yours and you’d rather it weren’t here? Open an issue.

unknown 2026-06-27 ♥414 ↻66 archive original ↗

"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." https://t.co/KkYE7vIr2R https://t.co/fg7iMlC9Nz

← back to the Pantheon