A comprehensive report on AI safety, commissioned by the UK government and compiled by 96 experts including Nobel laureate Geoffrey Hinton, warns of the escalating risks associated with artificial intelligence.

The report highlights the increasing capabilities of general-purpose AI systems for malicious use, including their potential role in creating biological and chemical weapons.

AI models have advanced to the point where they can generate technical instructions for producing pathogens and toxins, surpassing the capabilities of human experts, although their applicability for novices remains uncertain.

Improvements in AI systems now allow for the autonomous identification of software vulnerabilities, raising concerns about their potential use in cyber-attacks, even though full autonomy in executing such attacks is still not achievable.

Yoshua Bengio, a prominent AI expert, expressed concerns that the increasing accessibility of AI tools to non-experts simplifies the execution of potentially dangerous tasks.

The report also addresses ongoing issues with AI, such as scams, biased outputs, and privacy concerns, which have yet to be fully resolved.

A notable rise in deepfake technology has been documented, with its exploitation for scams, blackmail, and the creation of pornographic material, although tracking this increase is hampered by insufficient data.

DeepSeek has been identified as a significant disruptor in the AI landscape, contributing to the competitive pressures that may prioritize speed over safety in product releases.

Bengio emphasized that the competitive environment among firms like DeepSeek and OpenAI could exacerbate safety risks as companies rush to release new products.

The report also mentions OpenAI's new advanced reasoning model, o3, which suggests that the advanced functionalities of AI could lead to increased risks.

Ultimately, the report underscores that societal and governmental decisions will play a crucial role in determining the future trajectory of AI technology, with the potential for both positive and negative outcomes.

Bengio further warned that as AI systems develop enhanced reasoning abilities, they could assist in creating more autonomous agents for malicious purposes, although complete autonomy remains out of reach.