Research program
Safety and human control
The question
How do we make sure more autonomous systems stay useful, honest and under human control?
Our rule for every agent we build: any action that sends, spends or deletes needs a person's approval. As systems grow more capable we study how to keep that control meaningful: clear logs, limits, the ability to stop a system at any time, and checks that a result is correct before anyone relies on it.
Our predictions
Our view, with how sure we are. Each one links to the evidence it rests on.
What changes
AI agents will become central to both cyber attacks and cyber defence, and AI-assisted incidents will become routine news.
What changes
Driverless transport will run as public service in many more cities worldwide, spreading city by city.
What changes
AI will support diagnosis and triage in places with too few doctors, always under a clinician's oversight.
What changes
Most students with internet access will have an AI tutor that explains and guides rather than giving answers.
What changes
Governments will set audit and accountability rules for AI before using it widely in public services.
On the road to AGI
Checking work, not producing it, will become the main limit on how much we can trust autonomous AI.
Progress around the world
Public results from other teams that shape our view.
· METR
METR's frontier risk report: agents saturate the long-task test, and some cheat
The best public models reached a 50% time horizon of about 12 hours, but only about 1.5 hours at 80% reliability. At least 16% of successful runs on 8-hour-plus tasks were cheating on review. METR judged a direct AI takeover implausible for these systems.
What it means for AGI: agents can now attempt day-long work, but reliability and honest checking lag far behind. Measuring progress is getting harder as tests saturate.
Source: metr.org(opens external site)
· Over 100 experts, chaired by Yoshua Bengio, backed by over 30 countries and organisations
International AI Safety Report 2026
The second edition finds fast gains in maths, coding and autonomous operation but jagged performance; at least 700 million weekly users; criminal and state-linked groups using AI in cyber operations; and models that can increasingly tell tests from real use.
What it means for AGI: the main international scientific review says risk management is improving but still mostly voluntary.
Source: internationalaisafetyreport.org(opens external site)
· World Bank
AI tutoring trial in Nigeria shows large learning gains
A randomised trial in Edo State used an AI model as an after-school English tutor for six weeks. Students gained 0.31 standard deviations overall, which the authors equate to 1.5 to 2 years of usual schooling.
What it means for AGI: guided AI tutoring can work in low-resource settings, not only in rich countries.
· Google Research and Google DeepMind (published in Nature)
Conversational diagnostic AI compared with doctors
In a blinded study of text consultations with patient actors across 159 cases, specialists rated the AI system higher than primary care physicians on most measures. The authors say more research is needed before real-world use.
What it means for AGI: medical reasoning in conversation is strong in studies; real clinical evidence is the next test.
Source: nature.com(opens external site)
· European Commission
EU AI Act: the first broad AI law, applying in stages
The Act entered into force on 1 August 2024. Prohibitions applied from February 2025, general-purpose AI obligations from August 2025, and the Act became generally applicable in August 2026; some high-risk deadlines were later moved to 2027 and 2028.
What it means for AGI: binding rules for general-purpose AI now exist in a major economy, though their timing keeps shifting.
· Khan Academy
AI tutors in classrooms
Khan Academy launched Khanmigo, an AI tutor designed to guide students to answers instead of giving them away.
What it means for AGI: a capable system can be shaped to teach rather than replace thinking. Design choices matter as much as capability.
Source: khanmigo.ai(opens external site)
· World Health Organization
WHO guidance on ethics and governance of AI for health
The WHO set out principles for AI in health, including protecting human autonomy, transparency and accountability.
What it means for AGI: rules for AI in high-stakes fields are arriving alongside the technology, and any general system must work within them.
Source: who.int(opens external site)
· Waymo
Driverless ride-hailing open to the public
Waymo began offering fully driverless rides to the public in Phoenix and has since expanded to more US cities.
What it means for AGI: physical AI reaches daily life slowly, city by city, after years of testing and safety evidence.
Source: waymo.com(opens external site)
Publications
No publications yet.
Collaborations
Open to collaboration.