International AI Safety Report

[anssiko]

The International AI Safety Report was published in January 2025. This 300-page report assesses a wide range of risks posed by general-purpose AI and how to mitigate against them. The report also discusses the role of web browsers.

The first version is available at:

• https://www.gov.uk/government/publications/international-ai-safety-report-2025
• https://arxiv.org/abs/2501.17805

I'm opening this issue to gather insights from this report that could inform the Ethical Principles for Web Machine Learning document.

Cc @dontcallmedom @chrisn @torgo who might be interested.

[anssiko]

To bootstrap, one emergent theme in the report that intersects with the web is the rise of capable AI agents. An excepts from the report:

Increasingly capable AI agents – general-purpose AI systems that can autonomously act,
plan, and delegate to achieve goals – will likely present new, significant challenges for risk
management. AI agents typically work towards goals autonomously by using general
software such as web browsers and programming tools.

The web browser has traditionally acted as a user agent to facilitates end user interaction with web content. With the emergence of AI agents, the web browser is increasingly facilitating interaction between AI agents and web content. This should be an area of interest to the W3C community.

[dontcallmedom]

re AI Agent, see also the breakout I'm proposing to run in a couple of weeks w3c/breakouts-day-2025#7

[chrisn]

The report also discusses the role of web browsers.

I haven't had chance to read the report yet, but a quick search turns up the following mentions of browsers:

Page 22: "AI agents typically work towards goals autonomously by using general software such as web browsers and programming tools."

Page 203: "the ability to distinguish between genuine and AI-generated content can help to reduce the harmful use of generative models. For example, if web browsers were able to put reliability notices on content that was likely AI-generated, this would help to combat the spread of misinformation online." (which relates to the Authentic Web workshop series).