SafetySecurityTransparencycs.CY
International AI Safety Report 2025: Second Key Update: Technical Safeguards and Risk Management
Yoshua Bengio, Stephen Clare, Carina Prunkl, Maksym Andriushchenko, Ben Bucknall, Philip Fox, Nestor Maslej, Conor McGlynn, Malcolm Murray, Shalaleh Rismani, Stephen Casper, Jessica Newman, Daniel Privitera, Sören Mindermann, Daron Acemoglu, Thomas G. Dietterich, Fredrik Heintz, Geoffrey Hinton, Nick Jennings, Susan Leavy, Teresa Ludermir, Vidushi Marda, Helen Margetts, John McDermid, Jane Munga, Arvind Narayanan, Alondra Nelson, Clara Neppel, Gopal Ramchurn, Stuart Russell, Marietje Schaake, Bernhard Schölkopf, Alavaro Soto, Lee Tiedrich, Gaël Varoquaux, Andrew Yao, Ya-Qin Zhang, Leandro Aguirre, Olubunmi Ajala, Fahad Albalawi, Noora AlMalek, Christian Busch, André Carvalho, Jonathan Collas, Amandeep Gill, Ahmet Hatip, Juha Heikkilä, Chris Johnson, Gill Jolly, Ziv Katzir, Mary Kerema, Hiroaki Kitano, Antonio Krüger, Aoife McLysaght, Oleksii Molchanovskyi, Andrea Monti, Kyoung Mu Lee, Mona Nemer, Nuria Oliver, Raquel Pezoa, Audrey Plonk, José Portillo, Balaraman Ravindran, Hammam Riza, Crystal Rugege, Haroon Sheikh, Denise Wong, Yi Zeng, Liming Zhu
November 25, 2025
arXiv.org
7 citations
Key Takeaways
-
Biological weapons risk is now a concrete concern: Three leading AI developers have implemented enhanced safeguards on new models because their internal testing could not rule out the possibility that these models could be misused to help create biological weapons, indicating that AI safety risks have moved from theoretical to practical deployment decisions.
-
Technical safeguards are advancing across multiple approaches: The field is developing multiple complementary techniques to make AI safer, including adversarial training, improved data curation, and enhanced monitoring systems, suggesting that risk management is becoming more sophisticated and multi-layered.
-
Industry is formalizing safety commitments: The number of companies publishing Frontier AI Safety Frameworks more than doubled in 2025, indicating that responsible AI practices are becoming standardized business practices rather than optional add-ons.
-
Governance frameworks are emerging but remain limited: While governments and international organizations have established governance frameworks for general-purpose AI, these currently focus mainly on transparency and risk assessment rather than comprehensive regulation, suggesting the policy landscape is still developing.
-
Risk management requires coordination across stakeholders: The report shows that addressing AI safety involves researchers, public institutions, and AI developers working together on both technical safeguards and institutional frameworks, indicating that no single actor can manage these risks alone.