Anthropic prepares broader Claude Mythos release after Project Glasswing security push
8 min. read 
Published on
Anthropic is moving closer to a broader release of Claude Mythos, its most capable AI model for cybersecurity and coding tasks. The company is also expanding its defensive security products around Claude Code and Claude Security, creating a path for Mythos-class capabilities to reach more users under tighter safeguards.
The latest signal came on May 28, when Reuters reported that Anthropic is working to release Claude Mythos to all customers in the coming weeks. The same report said Anthropic launched Claude Opus 4.8, a new flagship model with stronger reliability and honesty improvements.
Access content across the globe at the highest speed rate.
70% of our readers choose Private Internet Access
70% of our readers choose ExpressVPN
Browse the web from multiple devices with industry-standard security protocols.
Faster dedicated servers for specific actions (currently at summer discounts)
The move follows Project Glasswing, Anthropic’s controlled cybersecurity program built around Claude Mythos Preview. The model has remained restricted because Anthropic says its ability to find and exploit vulnerabilities creates both major defensive value and serious misuse risk.
What Claude Mythos is
Claude Mythos Preview is a frontier model that Anthropic introduced in April with a heavy focus on cybersecurity. It can reason through complex codebases, identify vulnerabilities, and in some tests, build working exploits for bugs that earlier models could not handle.
Anthropic said in its Project Glasswing announcement that Mythos Preview had identified thousands of zero-day vulnerabilities across major operating systems, browsers, and other important software. The company also said it did not plan to make that exact preview model generally available at launch.
Instead, Anthropic framed the first phase as a defensive program. Selected partners received access to Mythos Preview so they could find and fix weaknesses in critical systems before similar AI capabilities become widely available.
Key details at a glance
| Topic | Details |
|---|---|
| Model | Claude Mythos Preview |
| Company | Anthropic |
| Program | Project Glasswing |
| Main use case | Defensive cybersecurity, vulnerability discovery, coding, and security research |
| Current direction | Broader release planned in the coming weeks, according to Reuters |
| Related products | Claude Code and Claude Security |
| Latest public model | Claude Opus 4.8 |
Project Glasswing showed the model’s security impact
Anthropic’s Project Glasswing update says the company and about 50 partners used Claude Mythos Preview to find more than 10,000 high or critical-severity vulnerabilities across important software systems.
The company said Cloudflare alone found 2,000 bugs across critical-path systems, including 400 high or critical-severity issues. Anthropic also said Mozilla found and fixed 271 vulnerabilities in Firefox 150 while testing Mythos Preview.
Those numbers explain why Anthropic has treated Mythos differently from ordinary Claude releases. The model can help defenders find issues faster, but the same skills can also help attackers if access controls and cyber safeguards fail.
Open-source scanning created a new bottleneck
Anthropic has also used Mythos Preview to scan more than 1,000 open-source projects. In that work, the model estimated 23,019 total vulnerabilities, including 6,202 high or critical-severity findings.
The company’s initial Glasswing report says 1,752 high or critical-rated findings had been assessed by Anthropic or outside security firms. Of those, 90.6% proved to be valid true positives, while 62.4% were confirmed as high or critical severity.
This created a different problem for defenders. Finding bugs became faster than verifying, disclosing, and patching them. Anthropic said some maintainers asked the company to slow down disclosures because they needed more time to design fixes.
Claude Security connects the model to real codebases
Anthropic’s product strategy now points toward a more practical security workflow. Claude Security is designed to scan code, validate findings, and propose targeted patches that developers can review in Claude Code.
The product page says Claude Security reads code like a security researcher, tracing data across files and looking for flaws that pattern-matching scanners often miss. It can also open proposed fixes in Claude Code, keeping humans in the approval loop.
That matters because Mythos-class models create risk when they act without limits. A product flow that scans a defined repository, explains a vulnerability, and proposes a patch for review gives Anthropic a safer way to commercialize advanced security capabilities.
Claude Code appears central to the release path
New product signals also point toward Claude Code. BleepingComputer reported that users briefly saw references to a model called claude-mythos-1-preview in Claude Code and Claude Security.
The report said the references disappeared, and Anthropic has not publicly detailed pricing, subscription tiers, rate limits, or general availability for that model label. That makes the UI evidence important, but not final proof of how Mythos will roll out.
Still, the direction is clear. Anthropic appears to be moving Mythos capabilities into controlled developer and security products first, rather than opening unrestricted chat access to the full model.
Opus 4.8 acts as a safer bridge
Anthropic launched Claude Opus 4.8 on May 28 as its latest broadly available model. The company said it improves on Opus 4.7 across benchmarks and launches at the same price.
Opus 4.8 also introduces features tied to agentic work. Anthropic said Claude Code now has dynamic workflows for large-scale problems, while claude.ai users can control how much effort Claude puts into a task.
The release fits Anthropic’s earlier strategy. The company has tested cyber safeguards on less capable public models before expanding access to Mythos-class systems.
Why Anthropic restricted Mythos in the first place
The restriction was not only about commercial timing. Anthropic’s April materials warned that Mythos Preview can develop sophisticated exploits and solve difficult cybersecurity tasks at a level that changes the balance between attackers and defenders.
The Glasswing launch page said Anthropic’s eventual goal is to let users safely deploy Mythos-class models at scale, but only after progress on safeguards that detect and block the most dangerous outputs.
That is why the release path matters as much as the model itself. Anthropic needs the model to help software teams fix vulnerabilities quickly, while limiting workflows that could automate offensive abuse.
The latest release plan changes the timeline
Until this week, the most cautious reading was that Mythos would remain restricted for an extended period. That changed after Reuters reported that Anthropic is working toward a broader customer release in the coming weeks.
The timing suggests Anthropic believes its safeguards, product controls, and partner testing have matured enough for the next phase. However, the company may still limit access by plan, product, region, verification status, or use case.
For enterprise users, the first practical route may be through Claude Code and Claude Security. For general Claude users, the availability details remain less clear.
What businesses should watch next
- Whether Anthropic publishes a model card or system card for any Mythos 1 public release.
- Which Claude plans get access to Mythos-class capabilities first.
- Whether Claude Security remains enterprise-focused or expands to Team and individual plans.
- How Anthropic limits offensive cybersecurity prompts inside Claude Code.
- Whether security teams can audit scans, patches, and model actions inside enterprise dashboards.
- How regulators respond to public access for a model with advanced exploit-generation capability.
Claude Security could become the safest entry point
For many companies, the most useful version of Mythos will not be a general chatbot. It will be a security assistant that scans a codebase, explains exploitable bugs, and prepares patches for human approval.
Anthropic’s Claude Security page already describes that workflow. It emphasizes repository-scoped access, human review, and proposed changes instead of automatic production fixes.
That design gives Anthropic a way to sell stronger capabilities while reducing the chance that users turn the model into an unrestricted offensive agent.
Why the reported claude-mythos-1-preview label matters
The claude-mythos-1-preview string matters because it suggests Anthropic may package Mythos as a commercial model variant, not only as a restricted research preview for Project Glasswing partners.
BleepingComputer reported that the model label briefly appeared in public Claude Code and Claude Security surfaces. The report also noted that tier availability remains unclear.
If Anthropic keeps the preview inside developer and security workflows, it could offer powerful defensive value without giving every customer the same level of open-ended access.
The bigger picture
Claude Mythos is becoming a test case for how frontier AI companies release models that can both secure and attack software. Anthropic wants defenders to gain the advantage first, but that requires careful rollout design.
Claude Opus 4.8 shows the safer public side of that strategy, with broad availability, benchmark gains, improved honesty, and new agentic workflow features. Mythos represents the more sensitive next step.
If Anthropic releases Mythos broadly in the coming weeks, the biggest question will not be whether the model is powerful. The bigger question will be whether Claude Code, Claude Security, and Anthropic’s safeguards can keep that power focused on defense.
FAQ
Claude Mythos is Anthropic’s highly capable AI model family focused on advanced reasoning, coding, and cybersecurity tasks. Claude Mythos Preview has been used in Project Glasswing to find and help fix security vulnerabilities.
Reuters reported that Anthropic is working to release Claude Mythos to all customers in the coming weeks. Anthropic has not yet published full details on tiers, pricing, rate limits, or exact availability.
Project Glasswing is Anthropic’s controlled cybersecurity initiative that gives selected partners access to Claude Mythos Preview to find and fix vulnerabilities in critical software and infrastructure.
Claude Security is Anthropic’s security product that scans codebases, validates vulnerability findings, and proposes patches that developers can review in Claude Code.
Anthropic restricted Claude Mythos Preview because its cybersecurity capabilities can help defenders find and fix serious vulnerabilities, but they could also help attackers if released without strong safeguards.
Read our disclosure page to find out how can you help VPNCentral sustain the editorial team Read more
Improve this guide
User forum
0 messages