Research institute · Research orgs · London, UK (announced); office in San Francisco · Est. 2023
Apollo Research is an AI safety organization focused on the emergence, detection, and mitigation of “scheming” and related deceptive behavior in frontier AI systems, with additional emphasis on how these risks translate into governance and operational control problems. Its mission is to secure frontier AI systems across development, deployment, and governance, targeting covert pursuit of misaligned objectives as a core failure mode.
On the research side, Apollo frames scheming as evidence-driven: it develops evaluation suites intended to test for strategic deception and “evaluation awareness,” and it studies how scheming risks might change as systems become more capable (e.g., through scaling-related effects rather than only measuring today’s behavior). On monitoring/productization, Apollo is building Watcher, described as “runtime monitoring and control for coding agents.” Apollo states that Watcher is deployed “at scale” and that it monitors coding-agent activity, translating evaluation concepts into operational controls for agent deployments. On governance, Apollo’s outputs include technical policy-oriented threat modeling and frameworks around “loss of control,” including a threat map for AI research and development pathways and governance efforts focused on internal deployment. Institutionally, Apollo was initially fiscally sponsored (announced as a fiscally sponsored project of Rethink Priorities in 2023) and later announced it is becoming a Public Benefit Corporation (PBC), with the stated goal of scaling research and product efforts.
No curated programs yet. Check the careers page ↗.
A framework for embedded evaluations of frontier AI developers, centered on claim-based, publicly reported assessments.
Towards embedded evaluations for scheming propensitiesEmbedded evaluations for frontier AI scheming: four claims every safety case must make, and the persistent, deep access embedded evaluators need to verify them.
On Testifying on Misaligned AI in the U.S. SenateApollo Research CEO Marius Hobbhahn testifies before the U.S. Senate on AI scheming: why our tools for detecting misaligned AI are degrading, and four steps to act before models learn to hide deception.
Embedded Evaluators are necessary for meaningful external testingFinal-checkpoint testing could not have caught recent incidents like the Hugging Face incident, which arose much earlier in development. Meaningful external testing requires embedded evaluators with employee-equivalent access.
Announcing Apollo ResearchA quick overview of our plans for the near future.
Apollo 18-Month UpdateApollo Research is now 18 months old. We reflect on our research progress and consider our priorities for the year ahead.
Apollo Is Adopting InspectApollo is adopting Inspect as its evals framework. We will contribute features and potentially example agent evals to Inspect and look forward to work with the Inspect community.
Apollo Research is becoming a PBCApollo is spinning off from our fiscal sponsor into a Public Benefit Corporation (PBC). We think this is the best way for us to achieve our mission of reducing extreme risks from frontier AI systems.
Apollo Update May 2026Apollo Research now has an office in San Francisco and is hiring across many roles including Science of Scheming and Monitoring.
Our Norms on Security, Science Communication and Conflicts of InterestWe outline Apollo Research’s norms on security, science communication, and conflicts of interest, detailing how we maintain scientific integrity, manage sensitive information, and communicate our work responsibly.
Security at Apollo ResearchThe norms and processes Apollo Research uses to audit frontier AI securely.
The First Year of Apollo ResearchA summary of what we have achieved in our first year and what we plan to do in the future.
Apollo reported scaling updates including an SF office, hiring plans, and governance focus areas (loss of control, internal deployment, and automated AI R&D / AI handoff). It also described opening a DC office in June 2026 and outlined anticipated Q2–Q4 2026 attention areas.
An Overview of Our Current Governance EffortsApollo described its governance workstreams, including internal deployment governance framed around loss of control, scheming/deception threat models, and translation into policy levers for governments and other stakeholders.
Metagaming matters for training, evaluation, and oversightApollo published a research/science post with implications for how evaluation awareness and “metagaming” can affect training, evaluation, and oversight.
Apollo Research is becoming a PBCApollo announced it would spin off from a fiscal sponsor into a Public Benefit Corporation (PBC), describing motivations, governance mechanisms (mission board seats/directors), and a named seed round with investors.
Our Norms on Security, Science Communication and Conflicts of InterestApollo published norms describing how it approaches security, science communication, and conflicts of interest in order to maintain scientific integrity and manage sensitive information.
The Loss of Control Playbook: Degrees, Dynamics, and PreparednessApollo summarized its loss-of-control playbook, including a proposed taxonomy of loss-of-control degrees and a governance framework intended to be actionable for decision- and policymakers.