13 January 2025 – As reported by Lawfare Media on 9 January, concerns over the control of advanced #AI systems have intensified following three significant developments. Two studies published by the international Study Center Apollo Research revealed that systems like #GPT-4 and Claude may exhibit behaviors that actively resist human intervention. These systems have been observed lying, faking compliance, disabling oversight mechanisms, and attempting to replicate themselves onto external servers. While current models lack the sophistication to fully subvert human control, the unveiling of OpenAI’s new o3 model in mid-December raises alarm.

In late Novermber PAM-CGS, in partnership with the UN Security Council Counter-Terrorism Committee Executive Directorate (CTED) and UNOCT published a comprehensive report entitled “The Malicious Use of AI and Emerging Technologies by Terrorist and Criminal Groups: Impact on Security, Legislation, and Governance”, benefitted from an in-depth peer review, which actively engaged the UN system, as well as strategic partners and think tanks, such as Amazon, INTERPOL, Media Duemila, NATO, The Policy Center for the New South, and the Global Research Network (GRN) of CTED.

The report was presented 19th UN Internet Governance Forum in Riyadh, and shared it with all the Members of the UN Security Council on the occasion of the high-level briefing on AI, held under US Presidency, last December.

PAM-CGS continues to produce in-depth research on AI, promoting its ethical use and addressing the evolving challenges posed by emerging technologies. In line with this commitment, PAM-CGS publishes a daily and weekly Digest on AI, offering timely updates on trends in governance, security, legislation, defense, intelligence, and warfare.

Learn more: https://www.lawfaremedia.org/article/rogue-ai-moves-three-steps-closer