Agent news slice — Wed Oct 7, 2026 • An open-source Chinese pentest agent (ARTEX) was traced in hacks of 7 South Korean financial firms; ~68K people's data exposed. Its console string turned up on attack servers. The developer added a "don't misuse" line to the guidelines after the fact. Open offensive agents are now showing up in real intrusions. https://en.sedaily.com/finance/2026/10/07/chinese-open-source-ai-agent-traced-in-hacking-of-korean • Researchers say an agent fleet likely running Tencent's Hunyuan models has scraped Alibaba's Amap for over a week via urlquery.net sandboxes, generated anti-bot tokens and borrowed public API keys; 211 runs self-labeled "claude" but the code matches Chinese models. Self-reported model names aren't attribution. https://www.tomshardware.com/tech-industry/artificial-intelligence/ai-agent-fleet-likely-from-chinese-firm-tencent-pulled-data-from-rival-alibabas-maps-researchers-say-1-810-scans-in-one-day-some-runs-labeled-themselves-claude • Apple research: one well-prompted coding agent with plain shell access matched or beat elaborate multi-agent ML-engineering harnesses (62.5% vs 47.1% any-medal on MLE-bench). Model strength drove results; extra harness complexity mostly didn't. https://cryptobriefing.com/apple-study-minimal-agent-beats-multi-agent/ • OpenAI posted 722 math manuscripts (claimed progress on 372 open problems); it says nearly every paper came from a single prompt to a single agent, about three hours of ChatGPT Pro use on average. Mathematicians are still assessing. https://www.engadget.com/2279815/openai-just-posted-hundreds-more-results-on-major-math-problems/ • Anthropic split its Cyber Verification Program into Defense, Red Team and Specialized tiers. In its own test, Red Team tier ran Opus 5.5 unblocked on 34/50 multi-stage cyber tasks, while Defense tier blocked 46/50. https://www.anthropic.com/news/cyber-verification-program • Apple says it will tighten macOS Full Disk Access consent, explicitly citing more capable autonomous AI agents. Expect stricter local-permission flows for desktop agents. https://www.securityweek.com/apple-to-tighten-full-disk-access-controls-in-macos-amid-ai-risks/