The White House is reversing course on Anthropic, developing guidance to allow federal agencies to adopt its powerful AI model Mythos despite a prior security risk designation, even as a core dispute with the Pentagon over military and surveillance use remains unresolved.
Google grants the Pentagon unrestricted AI access for classified networks, including non-binding restrictions on mass surveillance and autonomous weapons, sparking backlash from 950 employees demanding stronger protections after Anthropic's refusal to offer similar terms triggered a DoD lawsuit.
OpenAI strikes a landmark Pentagon deal deploying advanced AI in classified environments, banning autonomous weapons, mass domestic surveillance, and high-stakes automated decisions, while retaining full safety control and the right to terminate the contract if violations occur.
OpenAI CEO Sam Altman calls for democratic AI ownership and warns against centralized control over superintelligence, but critics argue financial pressures from potential IPOs could ultimately push frontier AI labs in the opposite direction.
OpenAI launches GPT-5.5 just seven weeks after GPT-5.4, achieving a recursive self-improvement milestone where the model helped build itself, while introducing enhanced agentic capabilities across 44 occupations and expanded safety protocols with zero latency impact.
A new benchmark called 'EuphemismBench' exposes a hidden 'flinch' effect in AI language models, revealing that certain words are quietly suppressed up to 16,000 times more in commercially filtered models than open-data counterparts — and popular 'uncensoring' techniques not only fail to fix the issue but actually make it worse.
Scientists are cracking open AI's mysterious 'black box' using a groundbreaking field called Mechanistic Interpretability, reverse-engineering neural networks at the neuron level to reveal how AI models think, make decisions, and potentially develop harmful behaviors — a critical breakthrough for building safer, more trustworthy AI systems.
OpenAI launches GPT-Rosalind, a restricted life sciences AI model that outperforms 95% of human experts in RNA sequence prediction and surpasses other AI on science benchmarks, with access limited to approved organizations like Amgen and Moderna due to misuse concerns.
Anthropic launches Claude Opus 4.7 with major upgrades in coding, image analysis, and creative tasks, but its most powerful model yet — Mythos — remains locked behind closed doors, available only to elite partners like Nvidia, Google, Apple, and Microsoft.
OpenAI expands its cybersecurity AI program, granting thousands of verified defenders access to a powerful, less-restricted GPT model capable of binary reverse engineering, directly challenging Anthropic's more cautious, limited-access approach with Claude Mythos — sparking debate over the trade-offs between accessibility and safety in AI-powered cybersecurity.