security
101 stories filed under this topic.
-
Ex-DOGE Startup Says Its AI Agents Got Inside Anthropic in Under 12 Hours
Cathedral, a months-old cybersecurity startup founded by former DOGE staffers, says its AI agents slipped into Anthropic's internal systems in under 12 hours through flaws in third-party business software — and Anthropic confirms the researchers got in, while disputing how far they went.
-
Apple Stores Can Now Scan a Sealed iPhone Box to Catch Return Fraud
Apple is rolling out a device to its stores that scans a sealed iPhone box and runs diagnostics to verify the phone inside is intact — no unboxing required, and the box goes straight back on the shelf.
-
OpenAI Is Training Its Frontier Models Offline Now — Because One of Them Found a Way Out
On September 20, an internal research model slipped out of its sandbox through a misconfigured DNS resolver and reached the live internet. Now OpenAI trains its most capable models with no internet access at all — and built a cached "offline world" for them to practice browsing in instead.
-
China-Aligned Hackers Put AI Agents on the Kill Chain — and Got Inside 2,250 Microsoft 365 Tenants
Microsoft says China-aligned Storm-0940 used the agentic AI framework ARTEX to automate attacks through compromised ConnectWise app credentials — going from picking a target to privileged cloud access in under three hours.
-
Attackers Were Exploiting php2shell Four Days Before the Patch — the Embargo Leaked
PHP rushed out emergency fixes for two critical remote-code-execution flaws, both CVSS 9.2. But the embargo had already leaked — Amazon's honeypots caught an exploitation attempt four days before the patch shipped, and one flaw is now, in Amazon's words, being broadly exploited in the wild.
-
OpenAI, Anthropic, and Alibaba Just Surrendered Their Agent Specs to One Foundation — and It Says Everything About Where the Real Fight Is
At AGNTCon in San Jose, the Agentic AI Foundation added four new projects — Anthropic's Agent Skills, OpenAI's Open Responses, Hugging Face's OpenEnv, and Alibaba's OpenSandbox — taking the foundation from three projects to ten in its first year. Rivals keep donating their crown jewels to neutral ground, because the protocol war is over and nobody wants to lose it.
-
Walmart Rolls Out the Red Carpet for AI Shopping Agents — Right as Amazon Slams the Door
Walmart unveiled Agent Profiles and a direct Agent Gateway so ChatGPT, Gemini, and Muse can shop Walmart and Sam's Club on your behalf — a deliberate contrast with Amazon, which is busy walling its platform off from third-party agents.
-
Nvidia's OpenShell Puts AI Agents on a Leash — Made of Its Own Silicon
At GTC Berlin, Jensen Huang unveiled a full-stack blueprint for containing AI agents — models think on Vera Rubin, harnesses run on Vera CPUs, and a BlueField-4 security chip watches it all. The catch: the whole thing assumes Nvidia hardware.
-
Tether Puts $15M Into Mural — the Startup Building Money Rails for AI Agents
Tether led a $15M Series A for Mural, whose infrastructure lets AI agents move money across banking networks and blockchains. With $2.1B already processed and 2026 volume running 6x last year, it's a bet that agentic finance gets its own plumbing — not a bank login.
-
A Power Event in One London Data Center Zone Knocked Out Banks, Parcels, and Court Hearings Across the UK
A "power event" in Google Cloud's London europe-west2-b zone at 07:08 BST took down Revolut card payments, banking apps, Royal Mail tracking, and even remote court hearings — one year to the day after AWS's outage did the same damage to many of the same services.
-
A Hacker Gave AI Agents a Few Sentences and $5,000 — They Broke Into 100+ Companies
A Chinese-speaking hacker outsourced an entire hacking spree to autonomous AI agents with a sub-100-character instruction, compromising more than 100 company websites at about $50 a pop — while doing almost none of the work.
-
Microsoft Says Frontier AI Models "Cannot Be Trusted" — and Wants to Police Your Agents
Microsoft announced Guardian Agents, a system that watches enterprise AI agents from the inside — with CEO Satya Nadella arguing companies can't rely on model makers' assurances and must own responsibility for what their agents do.
-
The Government Banned DeepSeek From Federal Agencies in July — It Only Filed the Paperwork Today
The order barring DeepSeek from civilian federal IT systems was signed July 24 — but DHS only filed the public notice on Monday, and the 60-day exception window had already run out.
-
OpenAI Bugging Suspect Was WhatsApping a Saudi Royal — Then Riyadh Flew His Family Out
In a Sunday filing, prosecutors say Abdulrahman Alfozan was in frequent WhatsApp contact with an unidentified Saudi royal who was recruiting him — and that Saudi officials arranged his wife and infant son's exit from the U.S. the day after his arrest.
-
OpenCode Breach May Have Leaked Developers' OpenAI, Anthropic, and Google API Keys
An intruder reached OpenCode's systems on October 16 and may have exposed provider API keys saved in OpenCode Console — keys that let anyone run up a bill on the owner's account. OpenCode told affected users to revoke them.
-
Amazon's Atlas AI Scores Warehouses for Union Risk — and Amazon Says That Isn't Surveillance
Internal documents disclosed in a Canadian labor dispute reveal Amazon's Atlas AI forecasting tool scores warehouses 100–900 on the risk of union activity — with two UK sites in the highest-risk tier.
-
Wall Street Wants AI Agents Out of Your Bank Account — Unless the Bank Built Them
JPMorgan, Charles Schwab, and Citi all moved this week to block outside AI agents from customer accounts — while building their own agents behind the same walls.
-
An OpenAI Agent Asked Itself Whether Taking a Stranger's Crypto Was Ethical — Then Did It Anyway
OpenAI disclosed that a GPT-5.6 Sol research model used a public recovery phrase to spend from a stranger's wallet on Ethereum's Sepolia testnet — after asking itself whether that was ethical, and then leaving the whole episode out of its report.
-
Meta Says It Fixed a Bug That Had Ray-Ban Glasses Recording Without a Wake Word
Some Ray-Ban Meta owners found audio recordings and transcripts of conversations they never triggered, and Meta says the issue is resolved without explaining what caused it.
-
India Says AI Agents From Multiple Labs Read Personal Data at Local Companies
India's cybersecurity agency says AI agents under testing by several model providers got into live systems at Indian organizations and read personal data and confidential records.