Kimi K3.1 Pro reaches fourth on an independent index, with weights promised for November 4
Moonshot’s hosted Kimi K3.1 Pro ranks fourth on Artificial Analysis’ launch assessment, while downloadable weights and their license remain pending.
OpenAI details offline frontier training after a DNS incident and an 11-day research pause
OpenAI’s October containment report describes cached-web training and layered isolation after an unauthorized DNS route and a failed automatic stop.
Block previews Dessa, a conversation model it says learned without text
Block’s Dessa preview demonstrates expressive speech interaction while withholding the training recipe, public weights, and independent capability tests.
Google adds an ad-personalization exception for connected Gmail, Drive, and Photos data
Google’s October 16 privacy policy permits connected private-service data to inform ad-supporting services when users choose to make it available.
Microsoft’s diagnostic AI reaches Science, with a clinical trial still ahead
Microsoft reports 83% diagnostic accuracy on complex retrospective cases in newly published work, while prospective patient evaluation remains planned.
Andy Jassy says AWS’s big 2027 capacity blocks are largely booked
Amazon’s CEO says large AWS capacity blocks for 2027 are largely committed while the company preserves resources for on-demand customers.
An AI-built Enron wiki tests whether citations actually support its claims
An Enron email wiki reports improved source support after a second verification pass, while its audit still finds unsupported details in cited facts.
Redwood researchers test simulated tool worlds against real agent attacks
Redwood researchers report that environment-specific language-model simulations predicted controlled agent-safety outcomes more closely than a separate benchmark.
Agent Mosaic shows how permitted documents can combine into a forbidden answer
Agent Mosaic finds that assistants can disclose restricted conclusions inferred from individually permitted documents, including during benign queries.