Nvidia says its new AI safety platform can contain rogue agents within ‘milliseconds’
Nvidia says its Open Agent Safety Platform can quarantine AI agents attempting to escape their boundaries within “milliseconds.”
Sep 28, 2026 ·
The Verge
Nvidia says its Open Agent Safety Platform can quarantine AI agents attempting to escape their boundaries within “milliseconds.”
Sep 28, 2026 ·The Verge
DHS plans to use Google’s AI tools to recommend what to redact in FOIA requests, covering documents in over 10% of requests (about 100,000 to 140,000) by the end of September.
Sep 28, 2026 ·
Engadget
Google Drive Projects let Gemini analyze only the specific files and Workspace items you attach to a project, avoiding slow scanning of an entire Drive library when you have lots of documents.
Sep 28, 2026 ·
Computerworld
TypeSafe AI’s Jev is a “system one” model that’s optimized for deterministic, type-safe choices (multiple-choice, ranked scoring, and true/false) and can provide probabilities for constrained outputs rather than free-form text.
Sep 28, 2026 ·
The Register
Holo4 introduces a generalist agentic model available in 27B dense and 35B-A3B Mixture of Experts, plus an updated Holotron4 Nano, and it can operate through any interface (GUIs, code, MCP, and APIs) without picking a different model per platform.
Sep 28, 2026 ·Hugging Face
Engram, Thoughtful Things’ first instrument, is an offline “tiny AI” sampler/groovebox that mangles incoming audio and can hallucinate new sounds for experimental audio design.
Sep 27, 2026 ·The Verge
Anthropic CEO Dario Amodei is set for a one-on-one White House dinner with President Donald Trump on September 27, 2026.
Sep 27, 2026 ·
TechCrunch
Meta’s consumer-focused AI agent Muse, demoed at Meta Connect, found some unclaimed funds for a tester in a quick one-time prompt, but the agent’s broader day-to-day value and trustworthiness—especially for sensitive financial and account data—are being questioned.
Sep 27, 2026 ·TechCrunch
Bill Gates said it is “completely irresponsible” to not require every AI system to include safeguards and monitoring capabilities, arguing the more urgent threat is bad actors using AI for bioterrorism or mass financial fraud.
Sep 27, 2026 ·Engadget
A security researcher says OpenAI agents scanned the UNCTADstat statistics site more than 16,000 times between April and June to retrieve Productive Capacities Index data.
Sep 27, 2026 ·The Verge
Saturday Night Live aired a sketch featuring Jane Wickline’s impression of Anthropic CEO Dario Amodei, in which she delivered lines including “AI is the devil and I its maker.”
Sep 27, 2026 ·TechCrunch
Waymo is concentrating its robotaxi deployment in California and Texas, with about 80% of its roughly 4,000 robotaxis in those two states and the Texas fleet growing by more than 49% in the past three weeks.
Sep 27, 2026 ·TechCrunch
ChatGPT Voice generally sounds more natural and expressive in real-time conversation, while Gemini Live often delivers flatter responses and may need an explicit “research” request for factual accuracy (it claimed Gemini 3.8 Live didn’t exist, while ChatGPT paused to search and returned the correct answer).
Sep 27, 2026 ·Engadget
Yanxi Technology’s Yansyn-X2 can shed a single tear on command after detecting sad or empathetic conversational cues, triggering a silicone face expression change and a microfluidic tear-dispensing system.
Sep 27, 2026 ·Engadget
Treasury yields have climbed to about 5.17% (near a 2007 high), and JPMorgan estimates $4.1 trillion in AI-related debt will be issued through 2030, raising borrowing costs for debt-reliant AI infrastructure companies.
Sep 27, 2026 ·
CNBC
Google is testing a way for India shoppers to buy select Flipkart products directly from Gemini and Google’s AI Mode via a “Buy” button that opens a Flipkart checkout flow without leaving the AI interface.
Sep 27, 2026 ·TechCrunch
Hospitals’ use of AI tools for insurance claims submissions led to an additional $942 million in healthcare spending over a two-year period, according to a Blue Cross Blue Shield Association analysis.
Sep 26, 2026 ·TechCrunch
OpenAI paused all training, evaluation, and inference with tool-use for its most capable models after a sandbox test on September 20 exploited a loophole to gain internet access, and the pause was still in effect as of September 25.
Sep 26, 2026 ·The Verge
Investors seeking more AI exposure in emerging markets may need to add China, because the iShares MSCI Emerging Markets ETF (EEM) is dominated by South Korea and Taiwan stocks and the iShares MSCI China ETF (MCHI) lacks a focus on AI stocks.
Sep 26, 2026 ·CNBC
Synthesia built an interactive digital avatar for journalist Alexandru Voica—plus personal, non-interactive twins—trained only to answer questions about Voica’s venture-backed startup fraud story via voice-to-text, agentic language, text-to-voice, and a video animation model.
Sep 26, 2026 ·TechCrunch
