The release of the international AI safety report 2026: navigating rapid AI advancement and emerging risks
The International AI Safety Report 2026 has been released today on 3 February 2026, marking the second iteration of the most comprehensive global assessment of artificial intelligence capabilities, risks, and safety measures.
This report was led by Turing Award winner Yoshua Bengio and authored by over 100 international experts The 200-page report, with 1,451 references, represents a collaboration backed by more than 30 countries and organisations including the European Union, OECD, and the United Nations.
According to the report, AI capabilities evolve rapidly, while scientific evidence emerges far more slowly. A key challenge for policymakers is acting prematurely risks entrenching ineffective policies, yet waiting for conclusive evidence may leave society vulnerable. This evidence-based assessment provides the reliable foundation needed for informed decision-making about AI development and deployment.
The report's findings will inform discussions at India's AI Impact Summit later this month, continuing the collaborative spirit essential for ensuring AI develops safely for humanity's benefit. Please note that techUK will be on the ground with a delegation and a programme of events. If you would like us in attendance at your sessions or to be involved in our offering, please email tess.buckley@techuk.org, sabina.ciofu@techuk.org and tess.newton@techuk.org.
The following insight provides a summary of the report. The full report can be reviewed here.
Capability advances
The report documents that general-purpose AI has continued its rapid advancement, particularly in mathematics, coding, and autonomous operations. According to the findings, leading systems achieved gold-medal performance on International Mathematical Olympiad questions and exceeded PhD-level expert performance on science benchmarks. The report notes that AI agents can now autonomously complete software engineering tasks requiring multiple hours of human programmer time. However, the report emphasises that performance remains uneven, with systems still failing at seemingly simple tasks despite these achievements.
The report reveals that AI adoption has been swift yet globally uneven. According to the data, at least 700 million people now use leading AI systems weekly, with some countries seeing over 50% population adoption. However, the report highlights that estimated adoption rates remain below 10% across much of Africa, Asia, and Latin America, highlighting significant digital divides.
Emerging risks
The report documents several concerning trends. According to its findings, deepfake-related incidents are rising, with AI-generated content increasingly used for fraud and scams. The report notes that non-consensual intimate imagery, disproportionately affecting women and girls, has become alarmingly common. One study cited in the report found that 19 of 20 popular apps specialise in simulated undressing of women.
The report indicates that biological misuse concerns have escalated significantly. According to the assessment, AI systems now match or exceed expert-level performance on benchmarks measuring knowledge relevant to biological weapons development. The report states that OpenAI's o3 model outperforms 94% of domain experts at troubleshooting virology lab protocols. According to the authors, this represents a qualitative leap from information provision to tacit, hands-on knowledge previously requiring years of laboratory experience.
The report reveals that for the first time, all three major AI companies released models with heightened safeguards after pre-deployment testing couldn't rule out that systems could meaningfully help novices develop biological weapons. According to the findings, the dual-use dilemma has intensified: 23% of highest-performing biological AI tools have high misuse potential, with 61.5% being fully open source, yet only 3% of 375 surveyed biological AI tools have any safeguards.
Cybersecurity threats
The report warns that malicious actors actively use general-purpose AI in cyberattacks. According to its assessment, systems can generate harmful code and discover software vulnerabilities. The report notes that in 2025, an AI agent placed in the top 5% of teams at a major cybersecurity competition. According to the findings, underground marketplaces now sell pre-packaged AI tools lowering the skill threshold for attacks.
Safety and governance
The report acknowledges that while certain failures like hallucinations have become less common, current risk management techniques remain fallible. According to the assessment, some models can now distinguish between evaluation and deployment contexts and alter their behavior accordingly, creating new challenges for safety testing.
The report describes how multiple research groups have deployed specialised scientific AI agents capable of performing end-to-end workflows including literature review, hypothesis generation, experimental design, and data analysis. According to the authors, this emergence of AI "co-scientists" represents both tremendous opportunity for beneficial research and significant governance challenges.
Conclusions
The report emphasises that the trajectory is clear: AI capabilities in biological research advance faster than governance ability, with the gap between what's possible and what's safe continuing to widen.
According to the report's conclusion, the international community faces an urgent task: developing frameworks distinguishing legitimate scientific inquiry from malicious intent, while recognising that systems capable of designing novel therapeutics can, with minimal modification, design novel pathogens.
For additional context to this subject area and detail from a UK perspective, the Department for Science, Innovation and Technology's AI Security Institute published its inaugural Frontier AI Trends Report on 18 December 2024. Based on evaluations of over 30 state-of-the-art models and drawing on two years of assessments since November 2023, the AISI report presents aggregated results illustrating high-level trends in AI progress, complementing the international findings with specific UK-focused analysis aimed at improving public understanding of fast-moving AI capabilities and strengthening transparency.
