Research Engineer, Frontier Safety Mitigations, DeepMind

Research Engineer, Frontier Safety Mitigations, DeepMind Overview

This role sits within the Frontier Safety Mitigation team in the Gemini Safety organization at DeepMind, based in London. The team’s mission is to reduce the risk of harmful model behavior as frontier AI systems become more capable, and to help ensure that advanced technology is deployed for broad public benefit. The work is highly collaborative and combines scientific research with practical engineering, with a strong emphasis on defense-in-depth safety measures for frontier models.

In this position, you will help lower launch risk for advanced models by protecting against misuse in areas such as cybersecurity, chemical, biological, radiological, nuclear, and conventional explosive threats, as well as harmful manipulation. The role involves building evaluation systems, carrying out red-teaming, researching and deploying mitigations both inside the model and outside it, and continuously monitoring for new risks. The team treats the possibility of dangerous capabilities seriously and sees proactive mitigation work as a core part of building safe AI.

What you will do

• Develop sophisticated classifiers and the data infrastructure behind them to detect misuse, taking responsibility for the process from automated evaluation through rapid iteration of the model.
• Create monitoring tools that operate across different contexts so the team can identify coordinated harmful behavior, including new methods for combining signals from multiple user sessions to uncover larger attack campaigns.
• Design and implement response systems at the account level that use data and partial automation to identify persistent malicious actors, follow their activity, and apply strikes based on strong production signals.
• Assess and secure agentic AI systems by defining threats, building test environments, and deploying mitigations against advanced agentic hacking techniques and long-horizon attacks.
• Advance research in automated red-teaming and adversarial robustness by using multi-turn and agentic attack strategies to systematically uncover misuse vulnerabilities.
• Contribute to the broader safety strategy for frontier models by helping build both model-internal and external defenses and by tracking emerging abuse patterns.
• Work as part of a team that values scientific rigor, collaboration, and shared responsibility for safety outcomes.

What the team is looking for

• A bachelor’s degree or equivalent practical experience.
• At least five years of software development experience across one or more programming languages.
• At least three years of experience launching, maintaining, or testing software products.
• At least one year of experience with software architecture and design.
• A PhD in Computer Science, Machine Learning, or a related area, or equivalent practical experience; alternatively, a strong publication record in venues such as NeurIPS, ICLR, ICML, or EMNLP.
• Experience in cybersecurity detection and response, including building large-scale classifiers and anomaly detection systems, and a track record of moving safety mitigations from research into production.
• Experience with adversarial machine learning, automated red-teaming, or model interpretability and probing.
• Experience contributing to or leading applied machine learning work, including LLM training, inference, and fine-tuning.
• Comfort using AI coding agents together with strong architectural judgment, plus hands-on experience with TPUs and JAX.
• Knowledge of AI control, chain-of-thought monitoring, monitorability, and related frontier safety research areas.
• Strong collaboration skills and the ability to work effectively in an interdisciplinary environment.
• English proficiency, since the company states that English is required for all roles unless the posting explicitly says otherwise.

Additional context

DeepMind describes itself as a scientific community focused on solving intelligence and using AI for widespread public benefit. The organization says it brings together interdisciplinary teams working on ambitious problems across multiple domains, with safety and ethics treated as top priorities. The posting also emphasizes that the company operates globally and offers opportunities to learn across different areas and career paths.

The role is listed in London, UK, and the page does not describe a remote or hybrid arrangement, so the safest reading is that it is an onsite London position.

How to apply and other notes

Applicants can use the Google Careers application process from the posting. The page also notes that information collected through a Google Careers profile and any submitted application is handled under Google’s Applicant and Candidate Privacy Policy.

Google states that it is an equal opportunity and affirmative action employer and that it is committed to building a workforce that reflects the users it serves. The company says it does not accept agency resumes and will not pay fees for unsolicited resumes. It also notes that equity grants, when offered, are discretionary and subject to Alphabet’s approval process and the terms of the relevant stock plan and grant agreement.

The posting does not include any statement about UK visa sponsorship or work authorization requirements.

Degree Requirement: Bachelor’s

Visa Sponsorship May be