AI Will Make Cyberattacks More Dangerous. Anthropic Has a New Plan to Protect Critical Infrastructure.

Anthropic is expanding its cybersecurity ambitions with a new initiative aimed at protecting some of America's most vulnerable infrastructure as artificial intelligence gives hackers increasingly sophisticated tools to launch attacks.

The company is launching its Critical Infrastructure Defense Program, bringing its most advanced models, cybersecurity researchers and engineers directly to companies responsible for protecting essential services, according to Axios.

Unlike traditional cybersecurity programs focused primarily on software networks, Anthropic's initiative targets systems that control physical operations, where a successful cyberattack could interrupt electricity, compromise water services or disrupt industrial production.

Under the program, participating cybersecurity companies will receive access Anthropic's frontier AI models, on-site engineering support and threat intelligence research. Those resources will help them identify security vulnerabilities in their customers' systems and develop ways to address them.

Anthropic told the outlet that several partners are already using Claude to identify and remediate vulnerabilities and assist their customers with similar work. The company is also introducing a separate initiative targeting a different but equally significant cybersecurity challenge of vulnerabilities in open-source software.

Its new OSS Scanner service will offer free, AI-powered security scans to participating open-source projects, periodically examining their code for potential weaknesses. Software maintainers will receive automated reports identifying vulnerabilities, explaining how attackers could exploit them and recommending possible fixes.

However, those reports will not undergo human review before delivery, raising the possibility of inaccurate findings that developers will need to verify independently. The announcement follows Anthropic's earlier Project Glasswing, which provided vetted organizations access to powerful AI models to uncover security flaws.

The company noted that the effort demonstrated AI's ability to discover vulnerabilities rapidly while exposing the fact that identifying weaknesses is often much easier than verifying and repairing them.

Anthropic's expansion also comes amid growing competition among AI developers to strengthen cybersecurity defenses. In September, OpenAI announced a $1 billion initiative to provide subsidized AI access, training and technical support to organizations protecting essential services, including electricity providers and water systems.

The urgency extends beyond competition between technology companies. In August, more than 100 organizations, including Anthropic, OpenAI and Microsoft, warned that critical infrastructure operators had limited time to prepare for increasingly capable AI-assisted cyberattacks.

Significant questions remain about the latest program. The company has not explained how its partners will safely test and implement AI-recommended fixes without potentially disrupting the infrastructure they are trying to protect. It also has not disclosed whether participating companies will receive free access to its models or who will pay the computing costs.