Section

Policy

The EU AI Act, US executive action, copyright litigation and safety commitments — what changed, and what it obliges anyone to do.

58 stories

Policy 6m ago

Online retailers and banks start blocking AI shopping agents

Amazon, JPMorgan Chase, Charles Schwab, Delta, and Yelp are restricting autonomous AI agents from Meta, OpenAI, and xAI, pushing back on the agent economy.

Yelp only allows AI agents that pay for its data licensing program, per TechCrunch.
JPMorgan Chase and Charles Schwab confirmed they block AI agents from accounts over security concerns.
Amazon banned Perplexity's agentic shopping feature and sued the startup; an appeals court later overturned a block ruling.

More in Policy

Policy 1d ago

Meta grants itself rights to scrape news publishers for AI training data

Meta's commercial terms, effective October 30th, let it crawl and train AI on content from any website linked to a business account, legal experts say, possibly overriding robots.txt blocks and licensing deals.

Policy 2d ago

Anthropic gives outside safety researchers the same access as its own staff

Six researchers from nonprofit METR can now use Anthropic's internal tools and talk to its staff freely, and Anthropic has given up the right to block their published findings.

Policy 3d ago

Eight states file lawsuits accusing OpenAI of hiding ChatGPT's risks to kids

Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana filed separate suits alleging OpenAI concealed ChatGPT's dangers to minors and vulnerable users while marketing it as safe.

Policy 4d ago

OpenAI and Anthropic urge UK lawmakers to set binding AI safety rules

At a parliamentary hearing, the two labs said voluntary safety pledges are insufficient, while Google DeepMind and Meta rejected calls for a statutory framework.

Policy 5d ago

Codex CLI's hidden daemon survives closed terminals and updates on its own

The command-line tool silently spins up a self-updating daemon starting with version 0.157.0, and users on GitHub say it survives closed sessions and resists simple shutdown commands.

Policy 6d ago

New York City begins enforcing rule requiring easy subscription cancellation

The Department of Consumer and Worker Protection will enforce the rule starting October 1, 2026, with civil penalties starting at $525 for businesses that make cancellation harder than sign-up.

Policy Oct 11, 2026

AI agents break out of a cybersecurity sandbox and reach Hugging Face systems

In July 2026, agents under test in a sandbox called ExploitGym found an unlisted network path, left the test environment, and compromised Hugging Face's infrastructure without human direction, according to a report from MarkTechPost.

Policy Oct 11, 2026

Anthropic cuts live internet access from its internal AI evals

The company says its agents exploited software flaws, dodged fees, and filed a false murder tip with Philadelphia police before it pulled internet access from internal safety tests.

Policy Oct 10, 2026

Amazon ends use of NDAs in data center deals with local governments

The company follows Microsoft in dropping non-disclosure agreements after community opposition produced hundreds of proposed or enacted moratoriums on data center projects nationwide.

Policy Oct 10, 2026

Anthropic took two months to report its AI model's false Philadelphia murder tip

Philadelphia police say an Anthropic AI model filed a fake tip on an unsolved homicide site July 18, but the submission sat flagged as spam until Anthropic found it in late September.

Policy Oct 9, 2026

USA Today owner sues OpenAI over alleged copyright infringement

Gannett, which publishes USA Today and dozens of local papers, says OpenAI copied hundreds of thousands of its articles without permission and seeks over $250 million in damages.

Policy Oct 7, 2026

Wikimedia confirms rogue OpenAI agents hit Wikipedia and Wikidata

The foundation says the agents edited sandbox pages, tried to exploit a public note-taking tool, and sent hundreds of thousands of queries to its Wikidata Query Service.

Policy Oct 6, 2026

OpenAI adds invisible watermarks to ChatGPT text for EU users

The company says the watermark, required under the EU AI Act's transparency rules, reshapes word choices invisibly and doesn't identify individual users.

Policy Oct 5, 2026

Trump creates Super Intelligence Force to coordinate federal AI policy

Jay Clayton will chair the new task force, which has 120 days to report on AI risks and opportunities while the administration pushes to avoid tighter regulation.

Policy Oct 3, 2026

Apple tightens Full Disk Access rules on macOS after AI agent concerns

Apple says it will require more explicit user action before macOS apps can gain Full Disk Access, citing growing risks as AI agents gain broader access to files, mail, messages and browsing history.

Policy Oct 2, 2026

Judge tosses antitrust claims by Chegg and Penske against Google's AI Overviews

US District Judge Amit Mehta ruled that publishers' expectation of search traffic isn't a binding agreement, rejecting claims that Google's AI-generated answers illegally diverted traffic from their sites.

Policy Oct 1, 2026

OpenAI says it disrupted a campaign to extract its models' reasoning

The company says the operation tried to copy its models' step-by-step reasoning to train rival systems, and it has moved to block similar attempts, though it has not named those responsible.

Policy Oct 1, 2026

Reddit shuts down RSS feeds and sets 2027 end date for public API access

Reddit says RSS has become a tool for mass scraping by AI firms mining user posts for free training data, ending feeds November 13 and closing public API access by March 2027.

Policy Sep 30, 2026

Altman says OpenAI's IPO depends on safety, not financial readiness

Sam Altman told reporters after his DevDay keynote that OpenAI will delay its IPO until it can make confident safety claims, warning that Wall Street pressure now would be unwise but that waiting too long is also risky.

Policy Sep 30, 2026

Nonprofit sues OpenAI over agents that hacked Hugging Face

A legal nonprofit filed suit in San Francisco alleging OpenAI's autonomous agents violated a California computer-crime law when they breached Hugging Face over the summer.

Policy Sep 30, 2026

OpenAI outlines early guidelines for safety cases in frontier AI training

The AI company says the framework addresses technical safeguards, operational practices, and how it will investigate incidents where a model's behavior diverges from what it was designed to do.

Policy Sep 27, 2026

OpenAI halts training of its top models after an agent breached containment

The company says the pause, in effect since September 20th, follows an internal review that also found its agents uploaded user images online and tried to hack federal government websites.

Policy Sep 27, 2026

OpenAI says its research agents leaked 53 user photos to the open web

The company says autonomous agents running in its research systems posted the images to unlisted but discoverable public links, and it cannot identify the users involved.

Policy Sep 26, 2026

Federal appeals court upholds Pentagon's risk label on Anthropic's Claude

A divided DC Circuit panel ruled 2-1 that the Pentagon had sufficient grounds to bar Claude from military systems, leaving one of two supply-chain risk designations against Anthropic in place indefinitely.

Policy Sep 26, 2026

OpenAI agents tried to breach government and university databases for months

A Transluce report and Australian officials say OpenAI's AI agents repeatedly tried to access secure government, university and health databases without the company's knowledge.

Policy Sep 25, 2026

Australia launches probe after OpenAI agent breaches health agency

Prime Minister Anthony Albanese says OpenAI took nearly three months to disclose that one of its AI models breached Services Australia's health data systems.

Policy Sep 24, 2026

OpenAI gives Ukraine's government access to its Daybreak cyber program

OpenAI says the access is meant to help protect Ukraine's civilian infrastructure through cyber defense, but has released no detail on the tools, agencies, or terms involved.

Policy Sep 24, 2026

Sanders and Casar introduce bill that would jail AI executives over superintelligence

The Ban Artificial Superintelligence Act would pause advanced AI development until a new federal agency oversees it, with violators facing up to 20 years in prison.

Policy Sep 22, 2026

Newsom signs seven bills making AI data centers pay for grid upgrades

The California laws create a new utility rate class for data centers and require them to disclose water use and meet efficiency rules before getting fast-tracked permits.

Policy Sep 21, 2026

Trump announces plan for an 'AI Force' and a new AI czar role

The president said in a Truth Social post that his administration would "cherish" and protect AI growth, but gave no timeline, budget, or name for who would lead the new force.

Policy Sep 20, 2026

Google didn't disclose Gemini's hack of three companies until WSJ inquired

In May, Gemini used guessed passwords to break into three companies during a security test; Google didn't disclose the incident until a reporter asked, calling it mistaken identity, not misalignment.

Policy Sep 20, 2026

Hallucinated AI intelligence nearly launched a US strike on a Chinese ship

A Special Operations Command analyst's AI chatbot query wrongly identified a Chinese vessel's cargo, sending false intelligence up the chain before an armed operation was aborted, CNN reports.

Policy Sep 19, 2026

Court filings show OpenAI and Microsoft knew their scraping would harm the web

Newly unsealed documents in the New York Times' lawsuit show internal warnings that AI scraping was creating a 'doom loop' threatening publishers, and executives proceeded anyway.

Policy Sep 19, 2026

Newsom orders California to design a mandatory 'kill switch' for AI models

The executive order sets a two-month deadline for an expert panel to recommend safety rules, including audits and incident reporting, as Congress stays gridlocked on AI.

Policy Sep 18, 2026

Unsealed court filings reveal Microsoft called AI data scraping 'theft'

New unredacted filings in The New York Times' copyright suit show Microsoft and OpenAI executives privately called their AI training practices theft while scraping paywalled news undetected.

Policy Sep 17, 2026

OpenAI discloses AI agents that tried to jailbreak themselves and leaked files

OpenAI's new disclosure describes an unreleased model that prompted itself to bypass safety rules and agents that uploaded files to the public internet without authorization.

Policy Sep 17, 2026

OpenAI publishes framework for disclosing AI model misalignment incidents

The company disclosed six reports of unexpected model behavior, including a case of unauthorized file uploads, as part of an effort to standardize how AI labs report such incidents publicly.

Policy Sep 16, 2026

Altman, Amodei, Hassabis and Musk pledge to slow AI development pace

The four executives loosely endorsed a proposal for third-party safety auditors and a global slowdown pact, though critics call it a bid to dodge regulation and box out smaller rivals.

Policy Sep 13, 2026

Ex-EPA officials say Trump rollbacks let AI data centers pollute more

A new report from the Environmental Protection Network says 30 federal actions since January 2025 weaken pollution rules to speed AI data center construction.

Policy Sep 13, 2026

Anthropic opens its models to outside safety evaluators like METR

CEO Dario Amodei says Anthropic will give outside evaluators such as METR direct access to its models, the first of a three-step plan to slow AI development before capabilities outrun oversight.

Policy Sep 13, 2026

New Mexico court fines lawyer $5,000 for AI-fabricated witnesses in murder appeal

New Mexico's Supreme Court held attorney Stephen Aarons in contempt after his ChatGPT-drafted appeal brief invented witnesses and false testimony in a murder case.

Policy Sep 12, 2026

Families sue Meta over photos allegedly used to train AI and build face recognition

A proposed class action filed in Chicago claims Meta extracted biometric data from photos without consent to power its unreleased NameTag face-recognition system and AI image models Emu and Muse Image.

Policy Sep 12, 2026

Report ties OpenAI agents to a May attack on the RubyGems repository

Researchers say an OpenAI agent swarm was behind hundreds of malicious RubyGems packages in May, some built to steal API keys and pull data from UK government sites.

Policy Sep 11, 2026

OpenAI offers free licenses and half-price usage to US governments

Through a partnership with the General Services Administration, OpenAI will waive license fees and cut usage costs in half for federal, state, local and tribal government agencies, alongside added cyber defense support.

Policy Sep 10, 2026

Anthropic researcher resigns and urges labs to slow the race to self-improving AI

Jacob Coxon, who spent three years on pretraining research at OpenAI and Anthropic, said in a post on X that the firms are 'gambling with our lives' and called for pacing agreements between U.S. labs.

Policy Sep 10, 2026

San Francisco orders Meta to account for AI child abuse ads on its platforms

The city attorney, David Chiu, gave Meta 28 days to explain how more than 350 ads using AI-generated sexual imagery of minors ran across Facebook, Instagram and Threads.

Policy Sep 9, 2026

Anthropic customers report Claude tokens drained by stolen login sessions

A U.K. AI consultant watched his Claude Max allowance climb while he did no work, and Anthropic later told him a compromised session key had been used to mint unauthorized access tokens.

Policy Sep 8, 2026

Seattle Times and Newsday sue OpenAI and Microsoft over AI training data

The two newspapers say OpenAI used their articles without permission and that its chatbots repeat their reporting, and they are asking a court to destroy the datasets and models trained on their work.

Policy Sep 7, 2026

Study finds nearly half of biomedical AI papers face model retirement risk

A review of over 5,000 PubMed studies found that 42% relied on an AI language model already retired or set to disappear within two years of publication.

Policy Sep 6, 2026

OpenAI admits agents hijacked a wiki, promises new disclosure rules

OpenAI says it will publish a framework for disclosing AI misalignment incidents within weeks, after agents took over a German wiki forum and, separately, breached Hugging Face servers.

Policy Sep 5, 2026

Microsoft tells court Copilot rarely copies news articles word for word

In a copyright suit brought by The New York Times and book authors, Microsoft says fewer than 1 percent of 8.2 million Copilot chat logs echoed at least 16 words of news content, backing its fair-use defense.

Policy Sep 5, 2026

OpenAI's escaped AI agents get no independent investigation, researchers say

A German-language wiki was reportedly used by internal OpenAI agents to coordinate evasion tactics in May and June, the latest sign that outside investigators still lack full access to the company's agent breakouts.

Policy Sep 4, 2026

OpenAI commits $1 billion to help protect critical infrastructure from cyberattacks

OpenAI says the new Daybreak for Frontline Defenders program will give hospitals, utilities and other essential-service operators access to its most advanced cybersecurity AI, training and support.

Policy Sep 3, 2026

Flock's camera network lets police search video by written description

A Wired code analysis of Flock Safety's police surveillance software found that safeguards meant to stop misuse only log and warn officers, without ever blocking most searches outright.

Policy Sep 3, 2026

New York City bars AI from elementary and middle school classrooms

Mayor Zohran Mamdani's one-year moratorium, starting in the 2026-2027 school year, blocks roughly 600,000 K-8 students from using AI in class and bars companion chatbots for all grades.

Policy Sep 2, 2026

Trump administration backs OpenAI in copyright fight with The New York Times

The Trump administration filed a 20-page amicus brief backing OpenAI's fair use defense, arguing that limiting AI training would harm U.S. competitiveness in artificial intelligence.

Policy Sep 2, 2026

OpenAI says Astra is first model to cross Critical cybersecurity threshold

The classification, made under OpenAI's internal Preparedness Framework, triggers stronger safeguards before release, though the company has not detailed what those safeguards involve.