When will Anthropic release Mythos publicly?

Anthropic committed on May 28, 2026 (alongside the Series H announcement and Opus 4.8 launch) that Mythos will reach 'wide release in coming weeks.' No exact date was published. Based on the public language and reporting from Fortune and Bloomberg, the most likely window is mid-to-late June 2026 through July 2026. Anthropic has not yet defined what 'wide release' means — it may be a research preview opened to all researchers, an enterprise tier, or a public API.

What is Anthropic Mythos?

Mythos is Anthropic's specialized cybersecurity AI model, designed to autonomously discover and exploit software vulnerabilities. In preview, Mythos has identified 23,000+ potential vulnerabilities across 1,000+ open-source projects, with 1,726 confirmed by external security firms (over 1,000 rated high or critical severity). Notable findings include a 17-year-old remote code execution bug in FreeBSD's NFS server (CVE-2026-4747) and a chained 4-vulnerability browser exploit escaping both renderer and OS sandboxes.

Who currently has access to Mythos?

As of June 2026, Mythos access is restricted to Anthropic's Project Glasswing — a defensive cybersecurity initiative that gives controlled access to select organizations. Confirmed Glasswing participants include Mozilla (Firefox vulnerability discovery), Palo Alto Networks, and the offensive security firm XBOW. Anthropic has not publicly disclosed the full Glasswing partner list. Wide release should significantly expand who can use Mythos, likely via an enterprise API tier with additional safeguards.

Will Mythos be available in the EU at launch?

Unclear. Anthropic withheld Mythos preview from the EU during the initial Glasswing rollout (cited concerns around the EU AI Act's dual-use technology provisions and offensive cyber capabilities). The wide release announcement did not address EU availability. Given the AI Act's classification of offensive cybersecurity AI as high-risk, expect Mythos to either ship to the EU with additional restrictions, ship later than the rest of the world, or remain US-only initially. Compare to GPT-5.5 Cyber, which had similar EU access issues in May 2026.

Quick Answer

When Does Anthropic Mythos Get a Wide Release? (June 2026)

Published: June 1, 2026

When Does Anthropic Mythos Get a Wide Release? (June 2026)

On May 28, 2026, alongside its $65B Series H announcement and the Claude Opus 4.8 launch, Anthropic publicly committed that Mythos will reach “wide release in coming weeks.” That’s the most concrete timing Anthropic has shared since Mythos first appeared in restricted preview in April 2026.

Here’s everything known about the release timeline, what wide release will likely mean, and what to expect.

Last verified: June 1, 2026.

TL;DR

Promised timing: “Wide release in coming weeks” (May 28, 2026 statement)
Most likely window: Mid-June through July 2026
What “wide release” means: Not yet defined — could be open research preview, enterprise API tier, or restricted public access
EU availability: Uncertain due to AI Act high-risk classification of offensive cyber AI
Current access: Project Glasswing only — Mozilla, Palo Alto Networks, XBOW, and a small set of undisclosed partners

What Anthropic actually said

The Fortune article from May 29, 2026 quotes Anthropic’s announcement: Mythos is “coming in weeks.” Bloomberg, NYT, and TechCrunch all paraphrase the same language: a wide release is imminent but not dated.

No model card, no public benchmark page, and no developer documentation for Mythos was published alongside the announcement. That’s a strong signal Anthropic is not yet ready to publish the technical details — which is what the safety review for offensive cyber capabilities would require.

What Mythos actually does

Mythos is Anthropic’s specialized cybersecurity AI model — not a general-purpose model. Built for autonomous vulnerability discovery and exploitation, with capabilities Anthropic describes as significantly beyond Claude Opus 4.8 on offensive cyber tasks.

Public results from the preview (via SecurityWeek and Anthropic’s Glasswing initial update):

23,000+ potential vulnerabilities found across 1,000+ open-source projects
1,726 vulnerabilities confirmed by external security firms
1,000+ rated high or critical severity
Anthropic’s estimate: total confirmed criticals could reach 3,900–6,200 as scans continue
Mozilla: 271 Firefox vulnerabilities identified with Mythos’s help
Palo Alto Networks: dozens of new vulnerabilities found
XBOW (autonomous offensive security): called Mythos “potent for vulnerability discovery”
Standout finding: a 17-year-old remote code execution vulnerability in FreeBSD’s NFS server (CVE-2026-4747)
Standout exploit: a browser exploit chaining 4 separate vulnerabilities to escape both renderer and OS sandboxes

These aren’t proof-of-concept results. These are real vulnerabilities in real software, which is exactly why Anthropic has been cautious about who gets access.

Why wide release is taking time

Three obvious reasons:

1. Safety review. Mythos is a dual-use technology — same capability that defends networks can attack them. Anthropic has been explicit in its public statements that broader release is gated on safeguards that prevent misuse. Those safeguards are still being built.

2. Liability and compliance. Public release of a vulnerability-discovery AI raises questions about negligence, disclosure norms, and platform liability. Anthropic’s legal team needs comfort that responsible disclosure workflows will hold up at scale.

3. The leaked-access incident. Per Tom’s Hardware (May 2026), unauthorized individuals briefly accessed Anthropic’s Mythos through a data breach involving third-party provider tools. That incident likely added a safety-review cycle to the wide release timeline.

What “wide release” probably means

Anthropic hasn’t defined the term. Three plausible interpretations:

Option A — Open research preview: a free-tier research access program, gated by application but with broad eligibility. Similar to how Anthropic released the original Claude Opus models. Most likely if Mythos is positioned as research infrastructure.

Option B — Enterprise API tier: a paid commercial API for security researchers and defenders, with KYC, contracts, and audit trails. Most likely if Anthropic positions Mythos as a defensive-security product.

Option C — Restricted but expanded Glasswing: Project Glasswing expands from its current ~10 participants to dozens or hundreds, but stays a closed program. Most likely if safety review concludes more access is needed but not unrestricted access.

Reading the tea leaves: Option B (enterprise API tier) fits the May 28 announcement timing alongside the Series H best. Anthropic raised $65B and needs to convert valuation into revenue. A Mythos enterprise tier with a high price point would be exactly that conversion.

What this means for security teams

If you run a security team and want to be early to Mythos:

Register interest now — Anthropic’s Glasswing page (anthropic.com/research/glasswing-initial-update) is the public-facing channel. Even before wide release, Anthropic has been adding Glasswing partners.
Prepare your responsible disclosure pipeline — wide release will dramatically increase the volume of vulnerabilities found in your software. You need to be able to triage and patch at scale.
Build a budget line item — if Mythos ships as an enterprise tier, expect pricing well above standard Claude API rates. Possibly tiered by scan volume or vulnerability count.

How Mythos compares to other offensive AI models

Model	Status (June 2026)	Public results
Anthropic Mythos	Wide release “in weeks”	23K+ vulns, 1.7K confirmed
OpenAI Daybreak (cyber model)	Limited preview	Comparable capabilities, fewer public results
GPT-5.5 Cyber	Limited US availability, EU withheld	Vulnerability discovery + remediation
DeepSeek V4 + offensive cyber tooling	Open weights	Used in some Chinese red-team research

Anthropic Mythos is currently the most-publicly-validated offensive cyber AI. That’s both the strongest argument for wide release (defensive value is huge) and the strongest argument for caution (misuse risk is huge).

Sources

Bottom line

Anthropic’s commitment is concrete: Mythos wide release in coming weeks. The exact date isn’t published. Best guess: mid-June to July 2026, likely as an enterprise API tier with KYC and audit requirements, possibly with EU restrictions. If you run a security team, start prepping your disclosure pipeline now.