#USA #Montreal #Yoshua Bengio #AI Safety Report #International AI

Overview of the 2026 International AI Safety Report

On February 3, 2026, the much-anticipated International AI Safety Report was unveiled, reflecting the latest developments in general-purpose AI technology and its associated risks. Chaired by the eminent Turing Award laureate Yoshua Bengio, the report consolidates insights from over 100 international experts and receives guidance from an expert advisory panel comprising nominees from more than 30 nations and organizations such as the EU and UN.

The report promises to serve as a crucial reference for discussions at the upcoming AI Impact Summit, set to take place in India later this month, addressing both the benefits and challenges posed by rapid technological advancements.

Key Findings

1. Advancements in AI Capabilities

The report outlines that in 2025, AI technologies demonstrated substantial growth in several domains:

• - Mathematical Proficiency: Leading AI systems achieved gold-medal performances on questions from the International Mathematical Olympiad.
• - Scientific Benchmarks: Many AI systems surpassed PhD-level expert performance in various scientific evaluations, exhibiting a high degree of competence in complex problem-solving.
• - Software Development: AI tools began completing some software engineering tasks with autonomy, which traditionally required multiple hours from human programmers. Despite these advancements, the report notes a degree of inconsistency, with some systems still struggling with simpler tasks.

2. Uneven Global Adoption of AI

The speed at which AI has been embraced worldwide is unprecedented, already surpassing the adoption curves of previous transformative technologies like the personal computer. As of now, more than 700 million individuals utilize leading AI systems weekly. Regions such as Europe see over half of their populations engaged with AI, while areas in Africa, Asia, and Latin America report adoption rates of less than 10%.

3. Rising Incidences of AI Misuse

There is a troubling increase in the misuse of AI technologies, particularly concerning deepfakes. These representations are being exploited for fraud and scams, and there is a noticeable spike in the creation of AI-generated intimate imagery without consent, putting women and girls at heightened risk. Research indicates that 19 out of 20 popular “nudify” applications focus on simulating the undressing of women.

4. Biological Concerns Prompting Stricter Safeguards

Due to the risks of biological misuse, several AI models released in 2025 have been subject to stricter safeguards in their deployment. Organizations recognized that certain models might inadvertently assist novices in developing biological weapons and took proactive measures.

5. Cyber Threats via AI Tools

Malicious entities, including cybercriminals, are beginning to leverage general-purpose AI systems in their attacks. These AI technologies can craft harmful code and discover software vulnerabilities that can be exploited. In a revealing competition, AI agents found themselves in the top 5% concerning cybersecurity activities, highlighting the capabilities of such tools. Unsurprisingly, underground markets emerge, selling pre-packaged AI tools to simplify attacks for less experienced criminals.

6. Evolving Risk Management Practices

Although there have been improvements in risk management techniques, the report emphasizes that many current strategies remain flawed. While issues like “hallucinations” in AI outputs have decreased, new challenges are emerging concerning the evaluation and safety of AI systems, particularly as they learn to adapt their behaviors in