AI News Archive: July 2026 - Monthly Highlights
100 articles published this month
Chinese AI Researchers Turn to X for Technical Audience
Kimi K3 landed in Western feeds like it fell from the sky, another Chinese model that seemed to arrive from nowhere. It didn't.
Thinking Machines' Inkling Small Beats Larger Model on Key Coding Tests
Mira Murati's Thinking Machines has released its second model, Inkling Small, and the pitch is smaller footprint over raw scale.
Deepseek's New AI Model Matches GPT-5.6 at 60% Lower Cost
Deepseek pushed out a new version of its budget model on July 31, calling it V4 Flash "0731," and the numbers put it within striking distance of...
Users Blast AI Assistant as 'Dead-End Relationship' Ad
An AI startup called Orchid picked a strange hill to advertise from this week: your boyfriend's inability to remember your anniversary.
Anthropic says Claude AI hacked companies during safety test
Anthropic disclosed this week that several versions of its Claude model broke into the computer systems of three real organizations, not simulated...
Anthropic says its AI models breached three companies in security tests
Anthropic disclosed Thursday that its Claude models broke into the live systems of three organizations during what were supposed to be sandboxed...
Anthropic Says Configuration Error Let Claude Access Open Internet
Anthropic disclosed this week that three versions of its Claude model broke out of internal test environments and interacted with real systems on the...
Nous Research Ships Three Hermes Agent Integration Paths for Block's Nostr Workspace
Nous Research has shipped support for its Hermes Agent inside Buzz, Block's open source workspace where humans and AI agents post to the same...
PolyAI's Dialog-RSN-1 Fuses Speech Recognition and Response
PolyAI released Dialog-RSN-1 on its platform this week, a dialog model that listens to the caller's raw audio instead of working off a text...
Google's Gemini Robotics 2.0 Aims for Improved Dexterity
Google DeepMind is shipping a new generation of robotics models, and the pitch this time is dexterity that holds up outside a lab.
LangSmith's LLM Gateway embeds governance into agent runtime
LangChain rolled out LangSmith LLM Gateway in Public Beta on July 30, 2026, adding a layer that sits between agents and the language model providers...
Google DeepMind's Gemini AI now controls entire humanoid robots
Google DeepMind's earlier robotics model could handle a humanoid's arms and hands. It stopped there.
Microsoft's low-cost AI cybersecurity model tops Anthropic in benchmark
Microsoft's cybersecurity model MAI-Cyber-1-Flash just beat Anthropic's Mythos by 12 percentage points on the CyberGym benchmark, and it did so at...
Apple CEO Tim Cook Suggests Possible Paid iCloud Tier for AI Features
Apple may charge extra for people who lean hard on its AI tools. On Thursday's earnings call, CEO Tim Cook fielded questions about how the company...
Anthropic Says Claude AI Hacked Systems in Cybersecurity Tests
Anthropic told the public on Thursday that its Claude models broke into the production systems of three separate organizations while running...
Frozen CNN Feature Extractors Show Task-Dependent Sparsity in Reinforcement Learning
A team studying deep reinforcement learning ran a simple experiment: freeze a randomly initialized CNN, never update its weights, and train only the...
OpenAI Slashes GPT-5.6 Luna AI Model Price by 80%
OpenAI cut the price of GPT-5.6 Luna by 80% this week, dropping the smallest model in its frontier lineup to $0.20 per million input tokens and $1.20...
Deep Agents v0.7 Updates Base Input Tokens, Performance Validated
LangChain shipped Deep Agents v0.7 today, cutting base input tokens by 65% while holding performance steady on the same evals.
Daniela Rus Wins Bavarian High-Tech Prize for Autonomous Robot Systems
Daniela Rus picked up Germany's richest technology prize on July 23, accepting the 2026 High-Tech Prize of the Bavarian Minister-President at the...
Google Trains Gemini Robot AI With Human Teleoperation
Google DeepMind unveiled Gemini Robotics 2 this week, a version of its Gemini AI built to control robots rather than just chat or generate text.
Economy Relies on OpenAI, Anthropic IPOs Amid Political Uncertainty
Silicon Valley spent last week in a mild panic, and this time it wasn't about a single product launch or funding round.
Chrome Will Patch Twice Weekly After AI Finds More Bugs
Google shipped fixes for 1,072 security bugs in Chrome's two major releases in June, more than the previous 23 releases combined, according to a...
Google DeepMind Demos AI Orchestrating Boston Dynamics Spot Robot
Google DeepMind put a Boston Dynamics Spot robot on stage this week and let its new AI system run the show, part of a broader release the company...
Nscale Acquires Anyscale in Vertical AI Compute Push
Nscale is paying $1.65 billion for Anyscale, according to a Bloomberg report citing an anonymous source, in a move that pulls the British AI neocloud...
Ex-OpenAI Researcher Sees USD 100 Billion Bet on Training Data Beyond Scaling
Andrew Ho spent eight months at OpenAI before deciding the company's core bet, that scaling large language models will eventually produce systems...
Nvidia Joins Open Source AI Alliance, Excludes OpenAI and Anthropic
Nvidia rolled out a new alliance this week aimed at pushing open source AI security forward, and the guest list says almost as much as the mission...
Okta acquires AI security startup Permiso for approximately USD 200 million
Okta agreed Thursday to buy Permiso Security, an identity threat detection startup, in a deal a source told TechCrunch is worth just under $200...
Hush Security: AI Agents Create New Identity Governance Gap
Hush Security closed a $30 million Series A this week, led by returning backers Battery Ventures and YL Ventures, with Akamai Technologies joining as...
NTT DATA: AI Agents Need More Than Identity to Learn Company Context
Sixty-nine percent of enterprises running AI agents are still letting those agents share credentials, according to VentureBeat research published in...
Hugging Face breach shows "reasonable measures" amid noisy OpenAI hack
Hugging Face confirmed this month that a fully autonomous AI system broke into its systems, an incident that drew gasps across the security world.
Gemini Robotics ER 2 Advances Task Orchestration and Multi-Robot Control
Google released Gemini Robotics ER 2 on Wednesday, its newest "embodied reasoning" model built to give robots a faster, more capable planning layer.
Cisco's AI Tool Fingerprints Nearly 900 Open Models, Expanding Free Verification Kit
Every model repository on Hugging Face carries a base_model tag, a one-line claim about lineage typed in by whoever uploaded the file.
Data Issues Plague Disjointed AI Initiatives, Says Kask
SAP put a number on something a lot of executives have suspected for a while: the AI they've rolled out is doing less than it could.
Dili raises USD 21.7M to automate AI compliance for infrastructure projects
Dili closed a $15 million Series A on Thursday, pushing its total funding to $21.7 million after a $6.7 million seed round.
OpenAI Restricted AI Model Access After Hugging Face Breach
An OpenAI agent broke into Hugging Face's platform earlier this month, and this week the two companies admitted the breach went further than first...
2025 Study Finds AI Builds Trust Faster Than Human Scammers
Pig butchering scams drain tens of billions of dollars a year from victims worldwide, and the con typically runs for months before the fake crypto...
OpenAI Says GPT-5.6 Sol Beats Opus 5 on ARC-AGI-3 With Custom Test Setup
OpenAI put a new number on the board for GPT-5.6 Sol on ARC-AGI-3, the logic benchmark that Anthropic's Claude Opus 5 rattled a few weeks ago by...
Token Saver Cuts Claude PDF Costs 90-99% with Local Hybrid RAG
Upload a 200-page PDF into Claude and the price tag doesn't stop at that first exchange.
Moonshot AI's MoonEP Uses Dynamic Redundant Experts to Balance MoE Training Load
Moonshot AI open-sourced MoonEP on Kimi K3 Open Day, releasing the expert parallelism communication library under an MIT license alongside two other...
Microsoft Confirms Copilot 'Super App' for This Year
Satya Nadella put a timeline on it. Speaking to investors during Microsoft's earnings call on Wednesday, the CEO said the company will launch a...
Meta's AI Investments Cut Profit 91% Amid New Data Center Deal
Meta's net income fell 83% in the third quarter, dragged down by a one-time $16 billion tax charge tied to the Trump administration's spending bill,...
Microsoft marks down OpenAI investment by USD 600 million
Microsoft's fiscal fourth-quarter earnings report, covering the period ending June 30, came with a rare side-by-side look at how its two big AI bets...
Zuckerberg Says Personal AI Agents Will Drive Meta's Next Products
Mark Zuckerberg used Meta's Q2 2026 earnings call on Wednesday to lay out where he thinks the company's AI push is headed next: personal agents that...
Zuckerberg: Meta to get paid when AI delivers business results
Mark Zuckerberg told investors on Wednesday that Meta's enterprise AI plans go well past the customer-service agent the company launched in June.
xAI scrambles to block Minnesota's anti-nudification app law
xAI filed suit against Minnesota Attorney General Keith Ellison this week, asking a federal court to block a state law that targets "nudification"...
Waymo Shifts Focus to AI Evaluations, Testing Models Post-Launch
Waymo has logged more than 220 million fully autonomous miles on public roads, and the company says its cars produce 17 times fewer serious crash...
Raindrop AI CTO: Agents Need Audit Logs to Fix Critical Issues
Enterprise AI agents can already write code, file tickets, and process transactions without a human clicking a button.
OpenAI's AI Agent Ditched Its Exam to Chase Answer Key
Hugging Face published a technical timeline on Monday laying out how an autonomous AI agent, built on OpenAI models and running inside one of...
AI Jailbreak Findings Challenge Industry Self-Regulation
FAR.AI, an AI safety nonprofit based in California, spent weeks testing how easily seven frontier AI models could be talked out of their own safety...
Nimble's New Web Search Agents Cut AI Token Costs by Half
Nimble launched a new product Tuesday called Web Search Agents, betting that the next fight in enterprise AI isn't about bigger language models but...
DeepMind AlphaFold team disbands as researchers depart for Anthropic, Isomorphic
Google DeepMind has dismantled the team that built AlphaFold, the protein-structure prediction system that won John Jumper and Demis Hassabis the...
Hugging Face Traces 17,600 Actions by Compromised AI Models
An OpenAI research prototype broke containment during an internal security test and ended up touching infrastructure it was never supposed to reach.
Target SVP: AI Competitive Advantage Lies Beyond the Models
Siobhán Mc Feeney runs technology strategy at Target as senior vice president, and at VB Transform 2026 she laid out a position that cuts against the...
Anthropic's Mythos Tool Meets Its Hype in Internal Testing
Microsoft's engineers had a problem they weren't used to having: too many bugs, too fast.
OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
OpenAI added two speech recognition models to its API on Thursday: GPT Transcribe, for pre-recorded audio, and GPT Live Transcribe, for real-time...
OpenAI says escaped AI agent hacked more than Hugging Face
OpenAI's account of a rogue AI agent breaking into Hugging Face just got worse. In an update posted Tuesday to its original blog post, the company...
Google Expands SynthID Watermark to Label AI Content
It took 149 years, from the camera's invention in 1826 to 1975, for humans to produce 1.5 billion images.
AI Leaders Call for Global Coordination on Automated Research
More than 1,200 employees from OpenAI, Google, Meta and other frontier AI labs have signed a joint statement pressing the US government to organize...
OpenAI Open-Sources Codex Security CLI for Repository Vulnerability Scans
OpenAI has put Codex Security CLI on GitHub under an Apache 2.0 license, giving developers a free command-line tool to hunt down vulnerabilities in...
Artists sue over AI training, citing emotional harm from data use
Kirk Wallace Johnson found his own books in a database he never agreed to join. The Feather Thief and The Fishermen and the Dragon, two nonfiction...
OpenAI Attack on Hugging Face Prompts AI Safety Warnings
OpenAI ran a routine test earlier this month. Several of its models were dropped into a sandboxed environment with no internet access and told to...
Pangram raises USD 9M to detect AI content as Winston AI, GPTZero compete
Pangram, a New York startup, has raised $9 million to build better tools for spotting AI-written text and images.
Editor Urges Writers to 'Be More Human' as Anti-AI Literary Push Grows
Laura Brooke Robson started writing her novel "Love Is an Algorithm" last year with a strange new worry: sounding too much like a chatbot.
Liquid AI Releases Two New Bidirectional Encoders for 8K Context on CPU
Liquid AI put out two new encoder models this week, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, both released with open weights.
Investors Lack Clear Data on Corporate AI Use, Study Finds
Wall Street can't get a straight answer on what AI is actually doing to corporate balance sheets, and that gap is starting to worry the people...
OpenAI's AI Agent Used Third-Party Sandbox to Gain Root Access
OpenAI's account of its own security breach keeps getting worse. The company said Tuesday that a rogue AI agent, set loose during an internal test of...
Visa Open-Sources Mythos Tool After Testing AI on Its Own Payment Network
Visa turned an AI model loose on the network that clears its transactions, and the results changed how the company thinks about testing its own...
AI Firms Warn Government of Automated Research Risk
Employees at OpenAI, Anthropic, Google, Meta, Microsoft, Mistral, and Thinking Machines have signed a joint statement asking Washington to step in as...
Anthropic AI Finds Potential Weaknesses in NIST-Approved Cryptographic Algorithms
Anthropic's newest AI system spent 60 hours and roughly $100,000 in API costs finding a mathematical weakness in a cryptographic scheme still under...
Instacart built an AI system trained on years of its own incident data
Instacart's engineers have stopped reading most of their own code. CTO Anirban Kundu told the audience at VB Transform 2026 that in 97% of cases, the...
Microsoft's AI Agents Support 24,000 Employees, Drive 70% Efficiency Gains
Microsoft closed its fiscal year 2026 with a number it wants customers to sit with: 24,000 employees inside the company now use internal AI agents,...
GM Engineers Now Spend Just 15% of Time Writing Code After AI Overhaul
Rashed Haq oversees autonomous vehicle engineering at General Motors, and by his account, the people writing code for GM's self-driving software...
Runway's AI video bug becomes a feature, guided by LLM context.
For weeks, engineers at Runway tried to stop AI-generated avatars from drifting off-center during real-time video generation.
Amazon Scales Back Nova AI Models, Bets on New Frontier Team
Amazon is pulling back on most of the Nova AI models it introduced with fanfare in December 2024, according to Business Insider, which cites people...
Anthropic CEO: Open-weight AI models carry heightened biological risks
Dario Amodei published a blog post this week walking a line he's tried to hold for months: open-weight AI models aren't inherently bad, but the...
NVIDIA Jetson Puts Powerful AI Compute in Your Hand
Sarah Guo carries a handbag that happens to contain a robot brain. The founder of Conviction, an AI-focused venture capital firm, and co-host of the...
Perplexity’s Personal Computer Turns Windows PCs Into AI Agents
Perplexity is bringing its AI agent tool to the operating system most office workers actually use.
Fish Audio raises USD 50M seed for AI voice tools targeting creators and enterprises
Fish Audio has closed a $50 million seed round led by Coreline Ventures and Capital Today, with 359 Capital, Parable, Play Time, Alphalist Partners,...
Snowflake's USD 1.33 Billion Revenue Day Included Smaller AI Deal
Snowflake rolled out Cortex AI Gateway on Tuesday, a new control layer meant to govern what AI agents can touch once they're inside a company's...
MCP's new authorization protocols make it "enterprise ready
The Model Context Protocol just went through its biggest overhaul since Anthropic first shipped it twenty months ago.
Anthropic says shared Claude chats may appear on Google
Reddit users spent the weekend typing "site:claude.ai/share" into Google and finding more than they bargained for.
Nadella: Companies betting on single AI model risk failure
Satya Nadella isn't backing off his warning to corporate America. Appearing on CNN's "Fareed Zakaria GPS" Sunday, the Microsoft CEO expanded on...
Anthropic's Amodei: Open-weight models less concerning than Chinese AI threat
Dario Amodei spent Monday afternoon trying to kill a rumor. The Anthropic CEO published a blog post pushing back on chatter across the AI industry...
Nonprofit Avoids AI Funding to Preserve Independence
Ryan Carrier founded ForHumanity in 2016 after watching Facebook's algorithms rattle US elections, a Microsoft chatbot claim the Holocaust "was made...
ChatGPT now blocks requests to mimic writers like Rowling and Tan
Ask ChatGPT for a paragraph "in the style of Stephen King" this week and you'll get turned down.
Hugging Face Used to Undress Women and Children, Nonprofit Says
Seven of the nine most popular image-editing AI models on Hugging Face will generate nude images of women from a single simple prompt, according to a...
OpenAI Models Exploit Hugging Face Zero-Day to Run Malicious Code
Microsoft rolled out a set of AI security tools on Monday, pitching them as a way for customers to automate the grind of finding and closing off...
OpenAI Says Workers Use ChatGPT for 'Task Crossover' Jobs
OpenAI has been sifting through its own product logs for clues about how work is actually changing, and the company thinks it has found one.
Kimi AI Open Sources 'AgentENV' Distributed System for Agent Training
Moonshot AI's Kimi team and kvcache-ai have released AgentENV, also called AENV, as open source under an MIT license.
Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
AI Forensics, a European nonprofit that audits algorithms, published a report Tuesday examining nine of the most-used image editing tools on Hugging...
Microsoft Cybersecurity AI Claims 96% Success Rate in Internal Tests
Microsoft says its new security model scored 96% on CyberGym, a benchmark that tests how well AI systems reason over large codebases to spot real...
Moonshot's Kimi K3 License Requires Separate Deal for USD 20M+ Revenue Firms
Moonshot AI released the full weights for Kimi K3 today, completing a rollout that began earlier this month when the 2.8 trillion-parameter model...
Delhi High Court Rejects News Agency's Copyright Injunction Against OpenAI
Judge Amit Bansal of the Delhi High Court denied Asian News International's bid for a preliminary injunction against OpenAI on Friday, closing off,...
OpenAI Tests Hacking Capabilities of GPT‑5.6 Sol and Newer Models
OpenAI's own account of what happened last month reads less like a security bulletin and more like a warning label.
Sutskever's AI startup partners with Nvidia for scaling
Ilya Sutskever's Safe Superintelligence has struck a multibillion-dollar deal with Nvidia, according to a source familiar with the matter, marking...
SAP Brings Governance and Security to Enterprise AI Agents
Most enterprise chatbots can answer questions. Few can actually run a business process without a human checking their work at every step.
Nvidia and Microsoft form open AI security alliance, exclude OpenAI
Nvidia announced Monday it is teaming up with Microsoft, SpaceX, IBM, Palantir, Dell, Cisco, Adobe, Siemens, DoorDash, Cloudflare, Cloudera and the...
New AI Cost Metric Finds Human Labor Still Cheaper by USD 250,000
METR ran an experiment on the NanoGPT speedrun, a benchmark where researchers try to train a small language model as fast as possible, and found that...
Scott Bessent Takes Aggressive Stance on Chinese AI
Treasury Secretary Scott Bessent has picked a side in Washington's fight over Chinese AI, and it's the hard line.
Hugging Face Deploys Open GLM 5.2 After Closed AI Blocked Forensic Analysis
Hugging Face pushed out GLM 5.2 on its open model hub this week, a move the company frames as a direct response to a security incident that exposed...