OpenAI's Joshua Achiam: Did We Already Reach AGI?

Frontier AI models now possess advanced cyber capabilities, enabling them to identify and exploit zero-day vulnerabilities, which presents both a powerful defensive tool and a profound security risk. State actors may leverage these capabilities for espionage or sabotage, potentially using data poisoning or complex jailbreaks to flip models against their own infrastructure. Long-term cyber defense will likely evolve into a high-compute game tree search, where the ability to allocate resources and think multiple moves ahead determines the outcome. Despite these rapid technological advancements, such as AI solving long-standing mathematical conjectures, public perception remains largely indifferent, reflecting a broader societal trend of normalizing profound shifts in the background of daily life. Joshua Achiam, outgoing Chief Futurist at OpenAI, highlights the necessity for robust testing and verification standards to address these emerging, exotic vulnerabilities in an increasingly metastable world.