Anthropic on October 8 announced Cyber Mission, which the company describes as a long-term commitment to give defenders tools, research and resources to protect systems such as power grids, water utilities and transportation. It has two priority areas: critical infrastructure and open-source software. The first maps to the newly created Critical Infrastructure Defense Program (CIDP); the second to OSS Scanner, which launched the same day and which we have covered separately.
The first 11 partners
CIDP launches with 11 partners: Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. Dragos, Nozomi Networks and Insane Cyber specialize in industrial control system and operational technology (OT) security, Rockwell Automation is itself an industrial automation equipment maker, and Hitachi is the only non-U.S. company on the list.
Anthropic is offering partners three things: access to frontier models, embedded engineers and threat research. Security vendors that serve critical infrastructure can register their interest. The earlier Project Glasswing has been folded into an expanded Cyber Verification Program, to which security teams of any size can apply.
Why the focus on operational technology
Anthropic's assessment in the announcement is that highly capable cybersecurity models are already widely available to attackers, while defensive tools have not yet reached enough defenders. The cost of exploiting vulnerabilities is falling, but verification, disclosure and remediation remain slow and manual. The company says that during Glasswing, going from discovering a vulnerability to fixing it often took months, while patch cycles for OT systems can stretch to a decade.
Operational technology refers to systems that directly control physical equipment: programmable logic controllers in power plants, SCADA in water treatment facilities, signaling equipment in rail transit. Patching these systems means scheduling downtime windows. Many devices carry safety certifications, so changing a single line of firmware can mean going through certification again. Legacy hardware that the vendor no longer supports is replaced on the equipment's own lifespan, often more than ten years. That is what Anthropic means by a decade-long cycle. Vendors such as Dragos and Nozomi spend their days monitoring for anomalies in environments where patching is not an option, and putting models in their hands is easier to put into practice than handing them straight to utilities.
The company also made a forecast: two years from now, AI will favor defenders. That is Anthropic's own judgment. The announcement offers no data to support it and does not say what "favor" is measured against.
From Glasswing to Cyber Mission
Putting together Anthropic's cybersecurity moves this year:
- On April 7, Project Glasswing launched with about 50 partners, including AWS, Apple, Cisco, CrowdStrike, Microsoft, NVIDIA and Palo Alto Networks, who were given access to a preview of Claude Mythos.
- Glasswing later extended Mythos access to about 150 critical infrastructure operators in more than 15 countries, spanning power, water, healthcare, telecommunications and hardware.
- In June, a cyber defense program for U.S. state, local, tribal and territorial governments went live, which the company says now covers more than half of U.S. states; we previously reported on California rolling out Claude to state government agencies at half price.
- In August, the Defender Advantage Fund was established to support pilot projects and to keep OSS Scanner free.
- On September 15, Rockwell Automation joined Glasswing.
- On October 8, Glasswing was folded into the Cyber Verification Program, and CIDP and OSS Scanner launched the same day.
Anthropic says Glasswing has helped partners find more than 10,000 high- or critical-severity vulnerabilities since launch. The figure comes from the company, and partners have not confirmed it individually. Palo Alto Networks also runs its own Frontier AI Critical Defense Program, with both Anthropic and OpenAI taking part, which applies "virtual patches" at the network layer; it is a separate program from CIDP.
The open-source half
Beyond OSS Scanner, the open-source track includes Claude for Open Source, which gives maintainers a free Claude Max subscription. Partners include the Python Software Foundation and the Apache Software Foundation, as well as Alpha-Omega and OpenSSF through the Linux Foundation; Akrites and Gold Eagle handle aggregating vulnerability reports.
Several questions go unanswered in the announcement: whether CIDP partners have to pay, how long the arrangement runs, and how operators' OT systems will connect to the models. None is addressed. None of the 11 partners is a Chinese company, and Anthropic's services are not offered in mainland China, so similar work on power and water systems there will have to rely on domestic vendors and domestic models. Anthropic says it will share what it learns, including what fails, and invites other AI companies, security firms and governments to join.
Sources: Anthropic official announcement, GovInfoSecurity, CocoLoop, Rockwell Automation press release, SDxCentral; the Anthropic announcement was used to verify the list of 11 partners and which program each initiative belongs to, and GovInfoSecurity to verify the scope of Glasswing.