Anthropic's OSS Scanner Sends Unreviewed AI Bug Reports

Anthropic launched OSS Scanner, a free opt-in service that sends open source maintainers AI-generated vulnerability reports without human review.

@sharedot

Anthropic's OSS Scanner Sends Unreviewed AI Bug Reports

By @sharedot · · 8 pages

  • Programming
  • Open Source
  • Security
  • AI

Anthropic launched OSS Scanner, a free opt-in service that sends open source maintainers AI-generated vulnerability reports without human review.

What happened: a free scanner that skips the reviewer

Anthropic launched OSS Scanner, a free, opt-in service that uses its strongest language models, including Claude Mythos, to automatically scan open source projects for security vulnerabilities.

According to Open Source For You, the offering grew out of Project Glasswing and gives maintainers two paths: traditional coordinated vulnerability disclosure with manual review, or the OSS fast-track scanner, which delivers fully model-generated reports instantly without human review.

The trade-off is explicit — faster and more frequent audits, but a risk of occasional invalid reports. Each report includes a proof of concept, an explanation, and a suggested fix where one exists, and Industrial Cyber reports Anthropic expects a true-positive rate above 90%.

Why it's surprising: maintainers asked for bulk unverified bugs

The counterintuitive twist is that the no-review pipeline wasn't imposed on the community — it was requested. Open Source For You reports that because exploits can now be developed in minutes, maintainers increasingly ask for bulk submissions of unverified findings to accelerate patching.

During Project Glasswing, some maintainers with the capacity to triage at scale asked for everything Anthropic's models found in their projects, reviewed or not, according to Industrial Cyber.

In an ecosystem where unsolicited low-quality bug reports have long been a maintainer burden, a segment of the open source world now prefers fast, noisy, machine-generated disclosure over slow, verified reports.

The evidence: 29,000 findings, only 6,000 triaged

The bottleneck that motivated the change is stark.

Open Source For You reports that AI-driven vulnerability detection has advanced rapidly: on the CyberGym benchmark, LLMs went from discovering under 20 per cent of vulnerabilities early last year to over 85 per cent currently, with significantly higher-quality bug reports.

Through recent scans of critical software, Anthropic's models identified over 29,000 potential vulnerabilities, but human review capacity created a bottleneck, allowing triage of only about 6,000 of them.

Industrial Cyber adds that vulnerabilities found during Project Glasswing sometimes remained unresolved for months, showing that finding bugs is no longer the hard part — verifying, prioritizing and fixing them is.

The stakes: open source underpins almost everything

Anthropic says almost all software relies on open source code, much of it maintained by small teams of volunteers, and that highly cyber-capable AI models are now widely available to attackers while defensive tools have not reached enough defenders.

Industrial Cyber notes the company frames the near term as unfavorable: the cost of exploiting vulnerabilities has dropped, while verifying, disclosing and fixing them remains slow and human-dependent.

Anthropic forecasts that in two years AI will favor defense, but until then services like OSS Scanner are aimed at shrinking the exploitable window on code that runs power grids, banks and everything else downstream.

Where it fits: OSS-Fuzz's heir, plus paid siblings

Open Source For You reports OSS Scanner is inspired by Google's OSS-Fuzz and complements Claude Security, Anthropic's enterprise code-scanning product, by bringing free, high-powered defensive capabilities directly to the open source community.

Industrial Cyber describes the surrounding program: the Defender Advantage Fund (0xDAF), launched in August, keeps OSS Scanner free; maintainers can also apply to Claude for Open Source for free Claude Max subscriptions and to the expanded Cyber Verification Program for broader access to Claude's cyber capabilities.

Anthropic says it has funded the Python Software Foundation, Alpha-Omega, OpenSSF and the Apache Software Foundation.

What comes next: automated patching and honest numbers

Anthropic describes OSS Scanner as a first step.

According to Industrial Cyber, the company plans to bring the scanner to more projects while keeping human-verified disclosure for those that need it, automate triage and patching — still mostly manual — for projects that want it, and research methods for hardening or rewriting code beyond patching.

It says it will proceed based on guidance from open source maintainers and foundations and share what works so other projects can replicate it.

Sources

  1. opensourceforu.com › Anthropic Debuts AI-Powered Security Tool For Open Source Code
  2. industrialcyber.co › Anthropic launches Critical Infrastructure Defense Program with security firms to protect industrial OT environments

More on Programming