Artificial Intelligence · Cybersecurity
Anthropic expands Project Glasswing
Anthropic said it was widening Project Glasswing to about 150 organisations in more than 15 countries, after earlier partners used Claude Mythos Preview to find more than 10,000 serious flaws.
- Anthropic said roughly 50 initial partners had access to Claude Mythos Preview from early April.
- It said those partners had since found more than 10,000 high- or critical-severity security flaws.
- Anthropic said it was extending Project Glasswing to about 150 new organisations in more than 15 countries after working with partners, the security industry, open-source maintainers and the US government.
- It said Claude Security used Claude Opus 4.8 to scan codebases and suggest patches, and that it would share some of its internal tools with trusted security teams.
Anthropic said Project Glasswing was its effort to help secure important software. It said about 50 initial partners had been using Claude Mythos Preview since early April to scan codebases for vulnerabilities.
The company said it was now adding about 150 new organisations, subject to security requirements. It said most of them operated in critical infrastructure sectors such as power, water, healthcare, communications and hardware.
Anthropic said many of the new partners were vendors whose code was relied on by organisations around the world, including governments. It said it estimated that a major attack on most partners could affect more than 100 million people.
The company said it had released Claude Security, which used Claude Opus 4.8 to scan code and suggest patches. It also said it planned to expand Project Glasswing further and scale up its Cyber Verification Program for more cyberdefence work.
Anthropic said it expected many other AI companies to have models with similar cyber capabilities within 6 to 12 months, and that some could be released without safeguards. It said that could make cyberattacks more frequent and less predictable.
Named in this story
Companies
- Anthropic
- announced the expansion of Project Glasswing and related tools
Governments and agencies
- US government
- was among the parties Anthropic said it had worked with
Products and systems
- Project Glasswing
- the collaborative effort being expanded to security partners
- Claude Mythos Preview
- the model partners used to scan codebases and write patches
- Claude Security
- the product Anthropic said could scan codebases and suggest patches
- Claude Opus 4.8
- the public frontier model used by Claude Security
- Cyber Verification Program
- the programme Anthropic said it intended to scale up
How the source tells it
The piece read as a company announcement, combining self-promotional claims about its models with warnings about future cyber risk.
- vendor boosterism It presented its own models and programme as the answer to broad industry security problems.
- alarm It framed successful attacks as catastrophic and said they could affect more than 100 million people.
- urgency It used language about an immediate need to adapt and about acting as quickly as possible.
- speculation presented as fact It forecast that many other AI companies would have similar models within 6 to 12 months and that attacks would become more frequent and less predictable.