AI News
AI news that matters. Updated daily.
No stories match your filters.
The Pentagon Used OpenAI Models Through Microsoft Before the Military Ban Lifted
According to a Wired report published March 5, 2026, sources allege the U.S. Defense Department experimented with OpenAI's models through Microsoft's Azure cloud platform while OpenAI still had an explicit ban on military applications in its usage policy.
DiligenceSquared Raises $5.9M to Replace Consultants with AI in M&A Research
DigilenceSquared, a startup founded by an ex-Blackstone principal and a former Boston Consulting Group consultant, has raised $5.9 million to build AI-powered M&A research tools. The company uses AI-moderated expert interviews and voice agents to conduct the kind of due diligence that private equity firms and corporate buyers typically outsource to expensive consulting firms.
Wired Podcast Covers AI's Growing Role in Defense and Middle East Conflict
Wired's "Uncanny Valley" podcast released an episode on March 5 covering the intersection of AI and military operations, with a focus on how AI companies have been deepening their ties with the Department of Defense during the ongoing conflict in the Middle East. The episode also touched on prediction market ethics and the streaming wars between Paramount and Netflix, but the AI-defense segment is the most relevant piece.
AWS Launches Amazon Connect Health, an AI Agent Platform for Healthcare
AWS announced Amazon Connect Health on March 5, 2026 - a purpose-built AI agent platform designed for healthcare providers. The platform extends AWS's existing Amazon Connect contact center service into clinical and administrative workflows.
US Reportedly Drafting Sweeping Chip Export Controls Covering All Countries
According to reports on March 5, 2026, the US government is considering a drafted proposal that would dramatically expand chip export controls. The key change: the government would play a role in approving every AI chip export sale, regardless of which country is buying.
ByteDance's Seedance 2.0 Hit a Wall: Compute Limits and Copyright Complaints
ByteDance launched Seedance 2.0 in early February 2026 as a direct competitor to OpenAI's Sora 2, Google's Veo 3.1, and Kuaishou's Kling 3.0. The model generates cinema-quality video with synchronized audio from text, image, or video inputs. It's available primarily through ByteDance's Jimeng AI app in mainland China, priced at roughly one yuan ($0.14) per second of generated video.
Luma Agents Ship: Multi-Model AI for Text, Image, Video, and Audio
Luma launched Luma Agents on March 5, 2026 - AI agents designed to handle end-to-end creative production across text, images, video, and audio. The system is powered by Uni-1, the first model in Luma's new "Unified Intelligence" family, which is a single multimodal reasoning system rather than separate models stitched together.
OpenAI Ships GPT-5.4 With Pro and Thinking Variants for Paid Users
OpenAI released GPT-5.4 on March 5, 2026, positioning it as "our most capable and efficient frontier model for professional work." The release includes two variants: a Pro version and a Thinking version, each targeting different use cases.
Google Explains How AI Mode Handles Visual Search Queries
Google published an explainer on March 5, 2026 detailing how AI Mode in Search processes visual queries - when you point your phone camera at something and ask a question about it. The post, part of their "Ask a Techspert" series, focuses on a technique called "query fan-out" that breaks down visual searches into multiple parallel processing paths.
Cursor Launches Automations: Agents That Trigger From Slack, Git, or Timers
Cursor, the AI-powered code editor from Anysphere, announced Automations on March 5, 2026. The feature lets developers set up AI coding agents that launch automatically based on external triggers - not just manual prompts.
Meta Sued Over AI Smart Glasses After Workers Reviewed Private User Footage
Meta is facing a lawsuit over privacy violations tied to its AI smart glasses. Lawyers filed the case on March 5, 2026, alleging that subcontractors working for Meta reviewed private footage captured by customers' glasses - including nudity, sexual content, and other sensitive material.
Anthropic's $200M Pentagon Deal Collapsed Over Military Access to Claude
Anthropic's $200 million contract with the U.S. Department of Defense has fallen apart. The core issue: the Pentagon wanted unrestricted access to Anthropic's AI, and Anthropic wasn't willing to hand that over.
Meta's AI Glasses Reportedly Send Intimate Footage to Human Reviewers in Kenya
Swedish outlets Svenska Dagbladet and Göteborgs-Posten published an investigation revealing that Meta's AI-powered smart glasses may be sending sensitive user footage to human reviewers based in Nairobi, Kenya.
Google's February Roundup: Gemini 3.1 Pro, Deep Think, and Nano Banana 2
Google packed a lot into February 2026. The headliner is Gemini 3.1 Pro, which Google says delivers more than double the reasoning performance of Gemini 3 Pro. It's rolling out across the Gemini app, API, and enterprise platforms, with higher usage limits for Google AI Pro and Ultra plan subscribers.
Netflix Acquires Ben Affleck's AI Post-Production Startup InterPositive
Netflix has acquired InterPositive, the AI filmmaking company founded by Ben Affleck. The deal puts an AI post-production tool inside one of the world's largest content producers.
Hugging Face and NXP Show How to Run Robot AI on a $40 Edge Chip
On March 5, 2026, NXP Semiconductors and Hugging Face published a detailed technical guide on deploying Vision-Language-Action (VLA) models for robotics on NXP's i.MX95 embedded processor. This isn't a research paper - it's a practical engineering guide covering dataset recording, model fine-tuning, and on-device optimization.
Study Shows AI Agents Can Link Anonymous Online Accounts to Real Identities
A newly published study confirms what privacy researchers have long feared: AI agents can now cross-reference writing patterns, posting habits, and behavioral signals to link anonymous online accounts back to their owners. The research looked at scenarios most people assume are safe - Reddit alt accounts, anonymous Glassdoor reviews, finsta pages, burner X accounts - and found that modern AI systems can identify connections between these accounts and a user's real identity with concerning accuracy.
OpenAI's CoT-Control Study: Reasoning Models Can't Hide Their Thinking
OpenAI published research on March 5, 2026 introducing CoT-Control, an open-source evaluation suite designed to answer a specific question: can reasoning models deliberately control or hide what's happening in their chain of thought?
GPT-5.4 System Card: First OpenAI Model With Cyber Mitigations Built In
Alongside the GPT-5.4 launch on March 5, 2026, OpenAI published the GPT-5.4 Thinking System Card detailing safety evaluations and mitigations for its new reasoning model.
Why LLMs Can't Verify What They Know - And Why That Matters for Security
Chris McCormick published an analysis on March 4, 2026, examining how LLMs process and "know" information compared to humans. The core argument: LLMs have a fundamentally different epistemic architecture that makes them inherently vulnerable to deception.
Grammarly Launches AI Writing Reviews Styled After Famous Authors
Grammarly is rolling out a new feature called "Expert AI Reviews" that offers writing feedback styled after well-known authors - including those who are no longer alive. The feature, reported by Wired on March 5, 2026, uses AI to simulate how specific authors might critique and comment on your writing.
OpenAI Launches Education Push With New Tools and Certifications
OpenAI announced on March 5, 2026 a new set of resources specifically targeting education. The package includes new tools for classroom use, certification programs, and measurement frameworks designed to help schools and universities close what they're calling "AI capability gaps."
Stripe's New Benchmark Tests If AI Agents Can Build Real Payment Integrations
Stripe published a benchmark on March 5, 2026, designed to test whether AI coding agents can build production-quality payment integrations from scratch. It's one of the first benchmarks focused on real-world API integration rather than isolated coding puzzles.
Killer-Skills Offers 2,500+ Free MCP Skills for Claude Code, Cursor, and Windsurf
Killer-Skills, an open-source marketplace for AI agent skills and MCP (Model Context Protocol) servers, surfaced on Hacker News this week. The platform lists over 2,500 skills across categories like research, coding, marketing, and analysis, with more than 50,000 cumulative downloads and 120+ active creators contributing.
Apple Music Launches Mandatory AI Transparency Tags for New Content
Apple Music introduced a metadata tagging system that requires record labels and music distributors to disclose when AI was used to create content on the platform. The system, called Transparency Tags, covers four categories: artwork, tracks (sound recordings), compositions (lyrics and musical elements), and music videos.
LLMs Can Identify Anonymous Users for $2 Each, ETH Zurich Study Finds
Researchers from ETH Zurich and Anthropic published a study showing that LLMs can strip anonymity from pseudonymous online accounts at scale, cheaply, and with high accuracy. The paper, titled "Large-scale online deanonymization with LLMs," tested whether AI agents could link anonymous accounts to real identities across platforms like Hacker News, Reddit, and LinkedIn.
Nvidia Pulls Back From OpenAI and Anthropic Investments, Huang's Reasoning Unclear
Nvidia CEO Jensen Huang said on March 4, 2026 that his company's investments in OpenAI and Anthropic will likely be its last direct stakes in AI labs. The statement came during public remarks, though the specifics of his reasoning were thin.
Seven Tech Giants Sign White House Pledge to Cover Data Center Power Costs
On March 4, 2026, leaders from Google, Meta, Microsoft, Oracle, OpenAI, Amazon, and xAI met at the White House to sign President Trump's "Ratepayer Protection Pledge." The deal responds to growing bipartisan concerns that AI data centers are driving up electricity costs for regular households.
AMD VP Used Claude Code to Build a Pure-Python GPU Driver
Anush Elangovan, AMD's VP of AI Software, used Claude Code to build a pure-Python user-space driver for AMD GPUs. The driver communicates directly with AMD hardware, bypassing the standard ROCm/HIP software stack. It's currently used for stress testing, debugging SDMA operations, and verifying compute/communications overlap.
ChatGPT for Excel Arrives With GPT-5.4 and Financial Data Integrations
OpenAI launched ChatGPT for Excel on March 5, 2026, alongside a set of financial data integrations. The product runs on GPT-5.4 and is designed to work directly inside Microsoft Excel for modeling, research, and analysis - with specific attention to regulated environments like finance and accounting.
OpenAI Publishes Five-Stage AI Adoption Framework for Enterprise Leaders
OpenAI published a new framework on March 5, 2026, outlining five AI value models that businesses should follow to build what it calls "durable business advantage." The framework maps a progression from basic workforce fluency - getting employees competent with AI tools - through to full process reinvention, where entire business workflows get rebuilt around AI capabilities.
OpenAI Launches Adoption Channel to Push Enterprise AI Playbooks
OpenAI launched a new "Adoption" content channel on its blog on March 5, 2026. The channel is dedicated to publishing practical insights and frameworks for turning AI progress into business results. It launched alongside the five AI value models framework, signaling that this isn't a one-off blog post but a sustained content effort aimed at enterprise decision-makers.