Applying cybersecurity lessons to keep future AI systems safe
The author argues that AI safety is at a familiar turning point similar to earlier privacy, cybersecurity, and online safety challenges. She recounts a child exploitation case where a public-private effort cut the time needed to provide investigators with usable leads from about a month to roughly a day. That experience suggests competing companies and government can cooperate when protecting people is the shared goal, and the same collaborative model should guide AI safety.
The account centers on a three-year-old given the pseudonym Adia, whose exploitative image appeared online. Investigators faced the challenge of locating her, determining which authorities had jurisdiction, and passing usable intelligence to officers empowered to act. Because the abuse technology crossed borders instantly while child-protection systems remained split among agencies, legal limits, and older procedures, the response could stretch to about a month.
A public-private effort called Intel Inside, Safer Children Outside, involving technology firms, nonprofits, and law enforcement, reportedly shortened that timeline to roughly one day. The author later joined the Global Council for Responsible AI, leading efforts on an international AI safety commission and red lines for autonomous systems.
If adopted, this collaborative model could affect AI developers, regulators, law enforcement, and people relying on AI in healthcare, finance, government, and infrastructure. Shared safety practices may reduce risks from excessive permissions, prompt injection, and compromised integrations, potentially making incidents less likely or less severe. However, coordination could also raise questions about oversight, accountability, and whether voluntary efforts move quickly enough. The story suggests that AI safety may depend less on one actor and more on sustained cooperation among competing organizations and public authorities.