
Anthropic Fast-Tracks AI Bug Reports to OSS Maintainers, Taps 11 Firms for OT Security
Anthropic launched OSS Scanner, a free opt-in service modeled on Google's OSS-Fuzz that uses its most capable Claude models to periodically scan open source projects and send maintainers model-generated, unreviewed vulnerability reports with PoCs and suggested fixes (targeting >90% true-positive rate). It also announced the Critical Infrastructure Defense Program (CIDP), pairing frontier Claude models and on-site engineers with eleven founding partners — Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation — to help OT/ICS operators in power, water, manufacturing and transportation remediate vulnerabilities that often can't be patched quickly. Both build on lessons from Anthropic's earlier Project Glasswing, which found many vulnerabilities but struggled to get them fixed in time.
What it is
OSS Scanner, Anthropic's free opt-in vulnerability-scanning service for open source projects, modeled on Google's OSS-Fuzz. It uses Anthropic's most capable Claude models to periodically scan projects maintainers have opted in, producing reports with a proof-of-concept exploit and, when available, a suggested fix.
How it works
Reports are model-generated and sent to maintainers without human review — this is the defining design choice. Anthropic built this after maintainers who could already triage findings at scale asked to receive everything the models found, including unreviewed output. Projects not opted into this unreviewed path still get Anthropic's standard human-verified coordinated vulnerability disclosure instead.
flowchart TD
A[Project scanned by Claude] --> B{Maintainer opted into OSS Scanner?}
B -->|Yes| C[Unreviewed report sent directly:<br/>PoC + possible fix]
B -->|No| D[Standard coordinated disclosure:<br/>human-verified]Preconditions / Affected
- Opt-in only — maintainers must explicitly enroll their project.
- Anthropic states the service is intended for projects with enough triage capacity to handle unreviewed findings at volume, not for all maintainers.
What to do
- Maintainers of actively-maintained OSS projects with spare triage capacity can opt in to get faster, pre-review access to Claude-identified vulnerabilities and PoCs.
- Expect a true-positive rate above 90%, per Anthropic's stated target — meaning up to roughly 1 in 10 reports may be inaccurate (e.g., wrong severity rating), so some manual filtering is still required even with opt-in.
- Projects that can't absorb unreviewed noise should rely on Anthropic's existing human-verified disclosure channel instead.
Caveats
Anthropic itself frames this as building on Project Glasswing, where many vulnerabilities were found but fixes often took months — the >90% true-positive figure is a target/expectation stated by Anthropic, not an independently verified measurement, and the page gives no data on report volume or false-positive handling in practice.
Technologies named here
- Open source software projects
- OSS Scanner
- Claude
- Operational Technology systems
- Project Glasswing
- Google's OSS-Fuzz
- Critical Infrastructure Defense Program
- Anthropic's coordinated vulnerability disclosure process
- Anthropic
- CrowdStrike
- OSS-Fuzz
- Dragos
- Nozomi Networks
- Palo Alto Networks
Threat actors named here
- Russian hackers
Related in AI Security
