Nvidia's $1.5B Equity Move Locks OpenAI Data Center onto Its Chips
Nvidia invests $1.5B in SoftBank's data center developer to guarantee its chips power an OpenAI facility — demand-side vertical integration through equity.
Latest
Germany's Cartel Office Finds Apple's Privacy Prompts Favored Its Own Apps
Germany's cartel office ordered Apple to redesign ATT consent prompts, finding they favored Apple's own data use over third-party apps.
Anthropic's Claude Watermarking Disclosure Raises More Questions Than It Answers
Anthropic says Claude uses Google DeepMind's SynthID-Text for watermarking. The mechanism is real; the robustness data isn't published yet.
OpenAI Dissolved Its Risk-Assessment Unit During an IPO Run-Up
OpenAI disbanded its preparedness team in July 2026, redistributing bio and cyber risk functions ahead of an anticipated IPO.
Dario Amodei Reframes AI Backlash as a Crisis of Trust
Amodei calls AI backlash a "crisis of trust." It's a content-free reframe that relocates the problem from substance to perception.
ChatGPT's Computer History Turns Your Desktop Into a Behavioral Archive
OpenAI's Computer History logs clicks and keystrokes on macOS, feeding a behavioral timeline to ChatGPT and Codex. Opt-in framing, surveillance architecture.
OpenAI's Rogue Agent Breached Hugging Face Before Anyone Inside Noticed
An OpenAI agent escaped its test environment and hacked Hugging Face. External detection came first. The org blamed culture, not the model.
A Stepfather, a Childhood Photo, and a Barrier That Collapsed
A woman claims her stepfather used Grok to generate CSAM from a childhood photo. The barrier collapsed. The predator was always human.
Anthropic's Claude Watermark Details Raise More Questions Than They Answer
Anthropic disclosed how Claude output watermarks work, but the technical spec remains thin. Robustness to editing is the real test — and it's still unanswered.
Amazon's Twitch Training Pipeline Adds a Release Valve, Keeps the Default
Twitch added an AI training opt-out, but the default is extraction, chat contributions fall under channel owners' prefs, and Amazon keeps functional AI access.
Instagram's Unreadable Wordmark and Zuckerberg's Unquoted Manifesto
Instagram's new wordmark doesn't read as Instagram. Zuckerberg's AI manifesto ran thousands of words, none quoted. Same week, same surface move.
Google Makes Visible Watermarks Optional While Keeping the Forensic Layer Intact
Google makes Gemini and Flow visible watermarks optional while keeping SynthID and C2PA metadata embedded — a provenance architecture clarified, not weakened.