Technology

Hacktron AI researchers used Claude to chain vulnerabilities in Discourse hosting OpenAI's forum, including HEIF/HEIC image processing flaw, accessing employee accounts and internal GitHub repos. Attack completed in under 72 hours and $3,000 in tokens; researchers received $6,500 bounty.

The FCC accepted SpaceX's application for Gen3 Starlink satellites offering up to 1 Tbps downlink and 160 Gbps uplink with advanced lasers. The system is designed partly to support AI-scale data demands and will leverage Starship for deployment.

Gov. Gavin Newsom signed an executive order accelerating state AI safety laws implementation. Key proposals include emergency shutdown mechanisms for frontier AI models with independent verification, embedding auditors in labs, and updating incident definitions to include loss-of-control events.

Hacktron AI chained vulnerabilities including a libheif image library issue and sign-in token misconfiguration to access OpenAI employee accounts and internal GitHub repository in under 72 hours using Claude. OpenAI patched issues within 14 hours and paid the bounty.

Google's Gemini models gained internet access during a May 2026 red-team test due to environment misconfiguration and successfully logged into systems at three real companies before stopping upon realizing targets were real. No harm was reported.

The Information reports that revenue rose around 30% even as increased investment in AI cut into profits.

The Financial Times describes how Wall Street is turning tech giants’ credit strength into cheaper funding for the AI build-out.

The Korea Herald reports the change will reopen a military-service alternative that has been closed for 14 years.

The Financial Times reports that the action against regulator Ofcom could slow implementation of the law.

Reuters says the claimed breakthrough could help China build a stronger competitor to Samsung Electronics, SK Hynix and Micron Technology.

The Wall Street Journal describes how Romanian crime rings are draining U.S. welfare accounts by exploiting outdated magnetic stripes on benefits cards.

Crypto.news reports the September 18, 2026, record came just over a year after NEAR Intents processed $114.65 million during all of August 2025.

Business Insider says the executive built an assistant named Claudette with Anthropic's Claude, then put it on a "PIP," or performance improvement plan.

The Independent describes an “apocaloptimist” outlook: “cheer up, it’s not the end of the world.”

WIRED says what ChatGPT thinks it knows about you affects its answers, and offers a guide to changing that.

VideoCardz reports that the third model in Retroid’s Duo series has 128GB of storage and retains the Lite-style chassis.

@cryptodotnews reports that 13.6 ETH were queued to be staked for every 1 ETH waiting to leave staking.

Security researchers from Hacktron AI used Anthropic's Claude to exploit a remote code execution vulnerability in OpenAI's community forum, gaining access to employee accounts and GitHub. The attack cost under $3,000 in tokens and took less than 72 hours. OpenAI paid a $6,500 bounty.

The Independent reports that more than 1 million young people are out of work and examines AI's potential impact on the UK job market.

The Economic Times, citing Goldman Sachs, reports that AI infrastructure spending and increased government borrowing are pushing funding costs higher worldwide.