Lead, Frontier Red Team (Cyber)

San Francisco, CA

About the Team

As part of Anthropic’s work on the frontier of cybersecurity, we are forming a new team focused on steering the world through the next generation of cybersecurity. It is now clear that frontier AI models – Mythos and beyond – will have dramatic impacts on cybersecurity. This team will research the impacts of advanced models, beyond Mythos, on security – and publish the research and build + deploy the defenses the world needs to navigate it smoothly. We believe many parts of cybersecurity will need to be fundamentally reimagined for the age of powerful models.

This team is incubated with the highest priority within the Frontier Red Team, which studies the catastrophic risks and builds and advocates for defenses. See our previous related work on Mythos, N-days, exploits, 0-day, and finding vulns in hardened codebase. We also run Project Glasswing.

About the Role

The Lead will oversee all of Anthropic’s mission-driven research on defending the world in an era of advanced cybersecurity capabilities. They will lead research to uncover the offensive and defensive capabilities of Claude, prototype defenses in the wild, inform how Claude is trained and safeguarded in order to favor defenders, and engage with the public and government. The Lead will, over the long run, focus on how to give defenders a permanent advantage. Done well, within a year we expect Mythos, Glasswing, and our vulnerability and exploit research to seem like a modest first step towards this vision. They will work with our Product, Training, Security, and Safeguards teams to support their missions.

The Lead will be responsible for all components of this vision. They will hire a world-class research team, lead them to produce frontier research on extremely quick timelines, maximize the public impact of that research, and build and deploy defenses into the world. The lead will develop a compelling vision for a dramatically more secure future, and communicate it to the company, industry, researchers, and government. The lead will also be responsible for informing and making difficult decisions at the highest levels of the company – how to broadly distribute access to advanced capabilities, how to safeguard the model, what threats to focus on, what systems to harden, and what high-risk, high-reward research to take bets on.

The role is based in San Francisco, but we expect some travel at least to Washington DC.

What You'll Do

• Be responsible for Anthropic’s overall defensive cybersecurity vision

• Hire and lead a world-class research team of AGI-pilled cybersecurity researchers and program managers

• Design an AGI cybersecurity research program that aims to secure the world

• Lead the team to execute on this research at the scale, speed, and quality of a frontier lab

• Work with other Anthropic teams, such as Training, Safeguards, and Policy, to address their biggest opportunities and challenges in Anthropic’s cybersecurity mission

• Identify field partners – maintainers, researchers, companies, governments – to do joint projects with

• Develop a strategy to publish and share research with the world for maximum impact; shut down and avoid low-impact research or research communication

• Translate technical findings into compelling demonstrations and artifacts that inform policymakers and the public

• Proactively identify critical strategic decisions for the company and the AI lab ecosystem, raise them, marshall resources to address them, and help the decisions be made as smoothly as possible

Sample Projects

• Writing the strategy for how a frontier AI lab can use frontier models and resources to most defend the world

• Building frameworks and tools that enable AI models to autonomously find and patch vulnerabilities using tens of trillions of tokens

• Running purple-team simulations where AI defenders compete against AI attackers in network environments

• Pointing autonomous AI systems at real-world security challenges (bug bounties, CTFs etc.) to characterize risks, defensive potential, and compare to human experts

• Building demonstrations of frontier AI cyber capabilities for policy stakeholders