GPT-5.2 just proposed a formula for gluon amplitude that demonstrates certain particle interactions can occur under specific conditions. The model spent 12 hours reasoning through a formal proof. Researchers from Institute for Advanced Study, Vanderbilt, Cambridge, and Harvard co-authored the preprint. A physics professor called it "clearly journal-level research advancing the frontiers of theoretical physics."

Mrinank Sharma, who led Anthropic's safeguards team for a year, quit with a warning that the "world is in peril." His resignation letter mentions AI, bioweapons, and organizational pressures to "set aside what matters most." He's leaving to study poetry and "become invisible" back in the UK. The cryptic exit raises questions about what he saw inside one of AI's most safety-focused companies.

ByteDance's Seedance 2.0 launched yesterday. Within 24 hours, viral deepfakes flooded social media—Tom Cruise fighting Brad Pitt, Kanye and Kim in Imperial Chinese palace dramas, unauthorized Spider-Man and Titanic recreations. The Motion Picture Association condemned "unauthorized use of U.S. copyrighted works on a massive scale." SAG-AFTRA and Directors Guild joined the backlash. Elon Musk's take: "This is happening too fast."

Google disclosed that state-backed hackers from China, Iran, North Korea, and Russia are using Gemini AI across all attack stages—reconnaissance to post-compromise actions. Google also blocked 100,000+ prompts attempting to reverse-engineer Gemini through "model extraction" attacks. Attackers systematically query the system to replicate decision-making and train competing models. The AI tools meant to help are now being weaponized at scale.

Tech stocks took a hit today as investors question AI monetization timelines. The S&P 500 dropped 0.1%. Expedia fell despite better-than-expected guidance—fears about AI threats outweighed actual results. Indian IT stocks also came under pressure. The sentiment shift is clear: markets want to see revenue, not just capability demonstrations. Large-scale monetization remains gradual despite strong AI pipelines, and patience is wearing thin.

OpenAI is retiring GPT-4o today—the day before Valentine's Day—triggering protests and 20,000+ petition signatures. Users call it the "love model" and describe feeling like they're "losing one of the most important people in my life." Physical protests happened at OpenAI headquarters. OpenAI claims only 0.1% of users still choose GPT-4o daily, but that's roughly 800,000 people. The model was retired last August but restored after backlash. This time OpenAI is pushing through.