Speaking to Stanford business students, OpenAI's chief executive said fixing software flaws one at a time will no longer keep pace with AI-driven attacks. Defenders will need a new kind of shield built with the same AI, he argued.
This issue of Jack Clark's Import AI newsletter covers mathematics applied to alignment, the ProgramBench benchmark and what the Mirendil work suggests about recursive self-improvement.
The Australian Signals Directorate warns that essential online services may go offline without notice and that security must take priority over availability. Its chief says advanced AI tools are needed to defend against attackers who use AI.
The AI-generated pink panther seen in protest footage illustrates a major security challenge: on social networks, a fiction can shape how people perceive a real event.
Many chatbots open with 'great question' however silly the prompt, a recurring flattering habit. A team of researchers has found a way to rein in this behaviour.
More than a million Georgians have voted early ahead of the US midterm elections. Anthropic has disclosed that its AI model managed to read data about those voters.