AI hunts bugs, attackers move faster, and regulators push back
6 stories~2 min read
- Top storySecurity
OpenSSH 'sshole' bug exploited hours after fix, giving root on Alpine Linux (opens in a new tab)
Previdian saw exploits of CVE-2026-123158 less than nine hours after disclosure, as attackers logged into passwordless accounts on Alpine and musl systems. If you run sshd in Alpine images or containers, update to OpenSSH 10.6p2 or your distro package and restart sshd now.
- AI
OpenAI releases GPT-6 Cyber, a security model it rates Critical for cyber risk (opens in a new tab)
In OpenAI's tests, the model found 81 high-severity Firefox bugs and built working exploits for 948 Linux kernel flaws. Access is limited to vetted Daybreak Red customers, but it shows how quickly AI bug hunting is shrinking patch windows.
- Regulation
Meta pauses business terms that would have let it crawl users' websites (opens in a new tab)
Meta shelved terms due 20 November after a Norwegian injunction, a Brazilian order with R$1M daily fines and a UK CMA probe. It shows courts and regulators can stop AI data grabs before they start, and your current terms with Meta stay as they are.
- Security
Claude Code caught sending users' email addresses to websites it visits (opens in a new tab)
At least seven bug reports, including GitHub issue #107757, show Anthropic's coding agent put account emails into web requests to sites like Wikipedia and the SEC. Update to 2.1.317, which asks before sending your account email, and check what agents send out.
- Regulation
US Justice Department launches task force to prosecute AI misuse (opens in a new tab)
The DOJ's new Super Intelligence Task Force will coordinate AI-related criminal and civil cases and take over challenges to state AI laws. It creates no new offence, but companies shipping AI should expect closer checks on their internal controls.
- AI
Thinking Machines launches Inkling 2 Small, a 594B-parameter AI model (opens in a new tab)
Mira Murati's lab released a mixture-of-experts model with 17 billion active parameters, built to automate AI training tasks. It gives teams another frontier-class option as Business Insider reports an Accel-led $5 billion round at a $40 billion pre-money valuation.