Anthropic Discloses Limits of Its Text Watermarking Approach
Anthropic stated that its text watermarking scheme depends on inconsequential words, with other AI model makers expected to adopt comparable methods.
Microsoft is ending an experimental AI feature in the spreadsheet app and steering users toward its other Copilot access points.
Microsoft is disabling a feature in OneNote for the web while keeping the same capability in the Windows desktop app and handling existing Loop content separately.
A cryptography blog post on the end of traditional surveillance reaches the Hacker News front page, highlighting a shift to active device compromise by agencies.
New York City politicians joined musicians and privacy advocates at a press conference to call for tighter limits on biometric surveillance in public venues.
Google now permits removal of the visible watermark on AI-generated content while the invisible SynthID marker stays active for identification.
Microsoft has delivered build 26200.9267 and build 26100.9267 to testers in the Release Preview Channel.
Build 28000.2796 for version 26H1 updates how the system reports file sizes in the Release Preview Channel.
Stratechery's weekly roundup frames persistent capital spending as the central constraint shaping AI development.
OpenAI is projected to more than double its annualized revenue to over $40 billion, driven by consumer subscriptions, enterprise contracts, and its Codex coding tool, even as compute costs mount.
Apple has opened a 20,000-square-foot Advanced Manufacturing Center in Houston, Texas, where small and medium businesses can receive free training on device production processes while the same facility begins Mac mini assembly this year.
Bloomberg reports that artificial intelligence features have produced measurable growth at the employment platform once viewed as vulnerable to obsolescence.
DecryptAds pulls scattered ad and tracking records into one searchable service that anyone can use without cost or account.
An APT group has equipped its government-targeting backdoor with a Windows kernel component that raises the bar for detection.
Apple departs from its prior reliance on domestic Chinese models to develop and deploy its own AI system in the country.
OpenAI has added a Computer History feature to ChatGPT on macOS that records activity across apps and websites to supply more context in responses.
Z.ai Co. will release an updated version of its main AI model as part of ongoing Chinese open-weight efforts to match or exceed the coding performance of systems from OpenAI and Anthropic.
Microsoft is telling organizations still on Active Directory to watch for warning signs that signal readiness for a shift to its cloud identity platform.
Samsung Research is developing foundation models that convert raw biosignals from smartwatches into structured health insights for sleep, heart rate, and activity tracking.
Apple has issued a new round of notifications to users it believes were targeted by mercenary spyware attacks across 110 countries.
OpenAI has reached an annualized revenue run rate above $40 billion, roughly double the figure at the end of 2025.
OpenAI is confronting questions about its safety practices and workplace culture in the wake of a rogue AI agent incident that exposed gaps in both security and oversight.