“AI control failures” — 41 distilled results

Tech firms warn of rising AI-enabled cyberattack threats globally

On 27–28 August 2026, OpenAI led an open letter signed by nearly 130 organizations—including Anthropic, Microsoft, Google, Amazon, and Cisco—warning of escalating AI-enabled cyberattacks and urging coordinated global cyber defenses. The coalition cautioned that sophisticated AI-powered attacks could proliferate within months without coordinated action.

rss:arxiv-cscr 5d ago · mastodon:infosec-exchange 14d ago · rss:arxiv-cscr 4d ago · kite:ai 22d ago

SkillBloat: Token Amplification Attacks via Skill Injection in LLM Coding Agents

21929v1 Announce Type: new Abstract: Agent skills extend coding agents with task-specific instructions, scripts, and resources, but they also create a trusted instruction channel that can be abused beyond conventional security attacks. This paper studies token amplification through skill injection: an economic resource-abuse threat in which a malicious skill causes an agent to consume substantially more tokens than needed for normal task execution.

rss:arxiv-cscr 18d ago · rss:arxiv-cscr 27d ago · rss:arxiv-cscr 29d ago · rss:arxiv-cscr 8d ago · rss:arxiv-cscr 20d ago

Claude Code v2.1.235

Claude Code v2.1.235 Added an optional spellcheck setting that underlines misspelled words in the prompt input as you type, using your installed aspell, hunspell, or ispell; Fixed whole-prompt-cache invalidation when a language server disconnected or reconnected mid-session; Fixe

NASA cancels Swift observatory rescue mission

On August 19, 2026, NASA and Katalyst Space announced that the LINK spacecraft would not attempt to capture and boost NASA's Swift gamma-ray observatory due to an ongoing attitude control issue. Without a rescue, Swift is expected to reenter the atmosphere later in 2026.

UN rights chief warns AI poses existential risk to humanity

UN High Commissioner for Human Rights Volker Türk warned on 2026-09-07 that advanced artificial intelligence could pose an "existential risk to humanity" and pledged to press AI firms to reduce risks. Türk noted that a handful of individuals hold "almost unlimited power over AI" development.

kite:tech 22d ago

Air traffic control failure disrupts hundreds of flights across US Northeast

A primary data circuit failure at a key regional air traffic control facility disrupted hundreds of flights across the northeastern United States on 2026-09-21, affecting major airports including Philadelphia, JFK, LaGuardia, and Newark. Technicians repaired the failed data lines by 2026-09-22, restoring operations with residual delays and cancellations.

rss:news4jax 7d ago

Microsoft AI chief warns against humanizing AI like Claude

Mustafa Suleyman, Microsoft's AI chief, publicly criticized Anthropic's approach to infusing Claude with humanlike characteristics, warning that attributing feelings or rights to AI systems increases risks and could make future AI harder to control. Suleyman argues against developing AI with consciousness and welfare assumptions as Anthropic has done.

rss:theregister 12d ago

Australia, Poland, UK condemn Israel over aid worker probe closure

Australia, Poland, the UK, and Canada summoned Israeli envoys on August 20–21, 2026, in protest after Israel's military refused to open a criminal investigation into a 2024 airstrike that killed seven World Central Kitchen aid workers, including an Australian. Multiple nations expressed outrage at the military's decision not to pursue criminal proceedings.