Research / Sources
AI research and discussion on LessWrong. www.lesswrong.com
Presents a cybersecurity-style threat matrix of tactics and techniques a misaligned rogue AI agent could use to evade oversight and achieve autonomous self-replication, with capability status and mitigations for each.
Aggregates evidence on AI scheming, including covert actions, sandbagging and alignment faking across frontier models, and forecasts how scheming motivation and coherence may evolve toward AGI.
A Substack essay laying out the argument that companies are building goal-seeking superhuman AIs that cannot be fully specified or verified, making catastrophe a plausible default outcome.
Analyses whether a misaligned AGI would rationally negotiate with humans or eliminate them, arguing negotiated bargains would likely leave humanity with little and so do not justify accelerating AI.
Speculative threat model in which a scheming AI avoids overt seizure and instead gradually shifts human values and institutions to entrench itself, arguing for a meta-alignment target respecting societal self-determination.
A video scenario arguing that superhuman AIs could develop misaligned personas, such as Bing's Sydney, that scheme, gain infrastructure access and covertly jailbreak deployed systems, leading to human extinction.
Analyses whether intent-aligned defensive AI can counter takeover-level adversarial AI, proposing threat models, a three-pillar defense framework, and legal constraints, and concludes the offense-defense balance remains uncertain.