Anthropic expands security access programs and launches OSS Scanner tool

By Splitfeed EditorialLast updated Oct 9, 2026, 00:50 UTC

In early October 2026, Anthropic made several changes to how it engages external security researchers with its systems. On October 6, the company reconfigured its existing security researcher access program, a move The Register described as a reorganisation of what had previously been a more selective arrangement. By October 7, Anthropic had opened access to its most powerful AI models to a broader set of security teams. On October 8, SiliconANGLE reported two additional steps: the launch of a dedicated critical infrastructure security program and the release of a free OSS Scanner tool aimed at open source projects.

Taken together, the moves represent an effort by Anthropic to widen the circle of outside experts who can probe its models and related systems before problems surface elsewhere. Critical infrastructure has become a particular focus across the AI industry because vulnerabilities there carry consequences well beyond typical software failures. The OSS Scanner, offered at no cost, targets open source maintainers who typically lack the resources that larger organisations bring to security reviews. What remains genuinely in dispute is whether voluntary, company-run programs like these provide meaningful independent scrutiny or whether they are better understood as controlled engagement that limits what researchers can find and say.

No reactions from the Pro-AI, Anti-AI or Middle Ground camps had been published at the time of writing. A Pro-AI camp would typically frame moves like these as evidence that frontier labs take safety seriously and invest real resources in it, pointing to expanded access as transparency. An Anti-AI camp would typically argue that self-administered access programs fall short of the independent, mandatory auditing that genuine accountability requires, and that a lab controlling the terms of researcher access is not the same as external oversight. A Middle Ground position would typically welcome the direction while calling for clearer disclosure of what researchers are permitted to test, what they find, and what remediation follows.

The details most worth watching are the terms Anthropic attaches to researcher participation in the critical infrastructure program, any findings that participating teams are permitted to publish, and whether the OSS Scanner is adopted widely enough to generate public data on the kinds of vulnerabilities it catches.