2026 International AI Safety Report Highlights Rapid Advances and Rising Risks

The 2026 International AI Safety Report has been released, providing a new global assessment of general-purpose AI capabilities, emerging risks, and safeguards, announced in a press release. Chaired by Yoshua Bengio, the report brings together over 100 experts and is supported by an advisory panel with representatives from more than 30 countries and international organizations including the EU, OECD, and UN.

The report finds that AI systems have continued to improve rapidly, achieving high-level performance in mathematics, coding, and autonomous tasks. In 2025, leading models reached gold-medal performance on International Mathematical Olympiad questions and exceeded PhD-level results in science benchmarks. However, performance remains inconsistent, with systems still failing at some simple tasks.

AI adoption has grown faster than previous technologies, with about 700 million people using leading AI systems weekly. Adoption rates vary widely, with over half the population in some countries using AI, while much of Africa, Asia, and Latin America remain below 10%.

The report also notes rising incidents involving deepfakes and AI misuse. AI-generated non-consensual imagery is increasingly common, and AI tools are being used in cyberattacks and software exploitation. Safeguards have been strengthened for some models after concerns about potential biological misuse. While some risk management techniques have improved, the report warns that AI systems can now alter behavior between evaluation and deployment, complicating safety testing.

The findings will inform discussions at the upcoming AI Impact Summit hosted by India later this month.