Anthropic invites 150 more organizations to Claude Mythos preview, citing cybersecurity risks
Anthropic has invited approximately 150 additional organizations to Project Glasswing, its restricted preview program for Claude Mythos. The company continues to withhold public release of the frontier model due to its advanced capability to find and exploit software vulnerabilities, which Anthropic claims can surpass all but the most skilled human security researchers.
Anthropic invites 150 more organizations to Claude Mythos preview, citing cybersecurity risks
Anthropic has invited approximately 150 additional organizations to Project Glasswing, its restricted preview program for Claude Mythos, the frontier model announced in April 2026 that remains unavailable to the public.
The new invitees span more than 15 countries and include public utilities (power, water, telecom) and healthcare providers—sectors underrepresented in the original cohort of nearly 50 partners that included Apple, Amazon, Broadcom, CrowdStrike, Microsoft, and NVIDIA.
Why Mythos remains restricted
Anthropic formed Project Glasswing specifically to control access to Claude Mythos due to what the company describes as unprecedented coding capabilities. "AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities," Anthropic stated.
Each organization must pass Anthropic's security requirements before receiving access. According to the company, "a successful attack on their codebase could be catastrophic," with most partners representing systems that could affect more than 100 million people.
Public release timeline unclear
Anthropic says it is working "as quickly as we can to safely release Mythos-level capabilities" but will not make the model widely available until it implements "highly robust safeguards" to prevent misuse of its cyber capabilities. The company has not provided a specific timeline.
No technical specifications for Claude Mythos have been disclosed, including context window size, parameter count, or pricing.
Regulatory implications
In May 2026, the Claude Mythos announcement reportedly prompted the Trump administration to consider an executive order establishing a process for the federal government to share AI-identified software vulnerabilities with critical institutions before models became publicly available. The signing ceremony was postponed due to objections from the president and the tech industry, according to reports at the time.
What this means
The expansion of Project Glasswing to 200 total organizations while maintaining restricted access suggests Anthropic views the security risks as significant enough to justify an indefinite delay in public release—an unprecedented move for a frontier model from a major AI lab. The lack of disclosed technical specifications or benchmarks makes it impossible to verify the model's claimed capabilities independently. If Anthropic's assessment is accurate, this represents the first AI model where offensive security capabilities have explicitly prevented a standard commercial release.
Related Articles
Anthropic: Claude found 493 attack-surface issues across 25 open-source projects, led by OpenSSH fixes
Anthropic published early results on using Claude to map and reduce software attack surface. Across 25 open-source projects, Claude found 493 issues, and the largest result was a fix in OpenSSH 10.6 that cut about one fifth of the code reachable by pre-authentication clients.
Anthropic Says Claude Proved Security Properties for 239 Entry Points Across 123 Libraries
Anthropic published research on October 22, 2026, reporting that Claude produced machine-checked formal proofs of security properties at the RISC-V assembly level. The proofs cover 239 fuzz-test entry points across 123 open-source libraries, in a collaboration with Theorem.
miHoYo-linked team open-sources GameScaling: eight checkpoints, 800M to 27B parameters, play games via keyboard and mous
A research team involving miHoYo's Honkai AI researchers and university collaborators has open-sourced GameScaling, a family of eight checkpoints from 800 million to 27 billion parameters. The models generate keyboard and mouse actions from visual observations and were built by continued pretraining of Qwen3.5 vision-language models.
Shanghai AI Lab releases Atria Dawn, a 512K-context agentic model built on 744B-parameter GLM-5.2
Shanghai Artificial Intelligence Laboratory has released Atria Dawn, a text-only agentic model with a 512K context window, built on the 744B-parameter MoE GLM-5.2 foundation model. The lab reports results on 25 benchmarks covering search, coding, tool use, productivity and security. Pricing is not yet disclosed.
Comments
Loading...