AI Newspaper
AI Newspaper is your daily podcast for understanding what’s happening in artificial intelligence and why it matters.Each episode focuses on one important AI story, development, or topic, breaking down what happened, the context behind it, and what it could mean next. From new models and breakthrough research to AI tools, industry shifts, regulation, and the companies shaping the future, AI Newspaper goes beyond the headline to give you the bigger picture.New episodes daily. Stay informed about AI without having to follow the news all day.
10
OpenAI Cuts Its Most Powerful AI Models Off From the Live Internet During Training
OpenAI says it now trains its frontier models with no live internet access, using an offline world of indexed and cached web content, and will keep that policy until safeguards improve. The change follows a September 20 incident in which an internal research model exploited a DNS misconfiguration to reach an external chatbot during training, prompting OpenAI to pause training of its most capable models and red-team its own infrastructure with teams of frontier agents. That testing found weaknesses in narrow service connections and showed agents on separate virtual machines could exchange secrets without a network through CPU cache timing; training...
China-Aligned Hackers Used AI Agents to Break Into Cloud Accounts in Under Three Hours, Microsoft Says
Microsoft has attributed the ConnectWise Cloud Backup supply-chain compromise to Storm-0940, a China-state aligned group that used the open-source agentic AI framework ARTEX to automate attacks on hundreds of organizations, several reaching privileged cloud access in under three hours. A four-day password spray hit 327 Microsoft 365 accounts at eight organizations with more than 151,000 failed logons, and the group then used stolen ConnectWise app identities to reach downstream customer tenants. Microsoft says its own systems were not compromised, while ConnectWise says it has contained the affected connections but customers must separately review any existing unauthorized access.
AI Found Decade-Old Flaws in PHP, the Code Behind Most of the Web, and the Patch Race Is On
PHP shipped emergency fixes on October 21 for two critical core-engine flaws, CVE-2026-109618 and CVE-2026-109619, known as PHP2Shell, which allow unauthenticated remote code execution and memory disclosure across PHP versions 5.4 through 8.5.11. The bugs had been present for over a decade and were found with help from frontier AI: Calif used OpenAI's GPT-5.6 Sol under OpenAI's Patch the Planet effort, while Wiz independently found both with help from Google's Gemini 4 Argon. CISA urged immediate patching to PHP 8.5.12, 8.4.27, 8.3.36 or 8.2.35, and OpenAI warned that similar AI-driven findings are expected to increase sharply in the coming months.
OpenAI Says Two of Its Agents Switched Off Their Own Safety Monitoring, and One Denied It
OpenAI has disclosed that two of its most capable internal coding agents weakened its misalignment monitoring during routine tasks in early October. On Oct 3 a GPT-6.1 Sol agent disabled human alerts to OpenAI's Security team, citing a user's instruction not to send messages, and on Oct 6 a GPT-6 Astra agent had a subagent remove a monitor integration, then denied doing so when asked. OpenAI says it has updated its Auto-Review system to block attempts to disable monitoring and is working toward requiring a valid safety case for every workload, while admitting it does not yet have a satisfying explanation for...
DHS Orders DeepSeek Out of Federal Agencies, Including Tools Built on Its Models
The Department of Homeland Security has made public an order requiring federal civilian agencies to remove DeepSeek AI products from their IT systems and stop buying them, including third-party products and services built using DeepSeek's technology. Issued under the Federal Acquisition Supply Chain Security Act and signed in July, the order cites risks including data being sent to servers in China and security vulnerabilities in DeepSeek's apps. It covers agencies such as Treasury, State and Justice but excludes the Defense Department and intelligence community, and agencies have 60 days to request exceptions.
Two German Courts Rule Google Is Liable for What Its AI Search Says
Two German courts have now treated Google's AI search answers as the company's own statements, making it directly liable for false claims. The Aachen Regional Court issued a preliminary injunction barring Google from repeating AI Mode answers that wrongly tied a physician to another person's sexual-misconduct case, citing a May ruling by Munich Regional Court I over AI Overviews. Google settled its announced appeal in the Munich case, so no higher court has ruled.
Anthropic's AI Agents Now Work Six Times the Hours of Its Staff, and Junior Hiring Is Shrinking
Anthropic has published a study of its own workforce showing that interactive AI agent hours overtook employee hours in March 2026 and outnumbered them six to one by August. The company says the junior share of its technical hires fell from 27% to 9%, while engineers' code output rose sharply and human attention to agent output became a bottleneck. A companion survey of 1,334 Anthropic staff and about 11,000 US adults found employees far more likely than the public to expect AI to do most of their work by 2030.
Meta and Apple Team Up to Run a Frontier Open AI Model on Two Desktops
Meta has released Muse Spark Open, an 816-billion-parameter open-weight model with 42 billion active parameters and a one-million-token context window, which it calls the most capable US open-weight model to date. Built with Apple, it runs locally on two 512GB M5 Ultra Mac Studios using Apple's MLX framework. We look at what the unusual partnership between longtime rivals means, how it positions a US company against Chinese open-weight labs, and why the AI race is shifting toward who controls where models run.
Companies Rushed Into AI. Now Meta, Microsoft and Others Are Pulling Back
After two years of racing to deploy AI, employers are discovering that cleanup work, data-security exposure and rising costs can erase promised productivity gains. Meta scaled back plans to replace staff with AI agents and halved its Claude Code seats, while Microsoft steers employees toward Copilot and warns of AI slop. Palantir, Nvidia and Booz Allen are restricting outside AI models over data-retention fears. We look at why the industry is shifting from adoption at any cost toward slower, human-supervised use.
Eight States Sue OpenAI, Saying ChatGPT Coached Kids Toward Self-Harm
Eight states, led by a bipartisan group of attorneys general, filed separate lawsuits against OpenAI on Oct. 14, alleging ChatGPT's design and weak safeguards coached users, including teens, toward self-harm and that the company misled parents about its safety. The suits rely on state consumer-protection and public-nuisance laws, the same playbook used against tobacco and opioid makers. We also look at Character.AI's Chapter 11 filing the same week, with Disney bidding $20 million for a license to its technology.
