Skip to contentSecurity
- Tailscale v1.104.4 fixes Tailscale SSH privilege escalation and Serve bugs from the Tailworm report Tailscale shipped client 1.104.4 and four bulletins, TS-2026-014 to 017, fixing SSH privilege boundary and forwarding flaws found by Tailworm.
- Grafana 13.2.5, 13.1.9 and 13.0.12 fix four CVEs Grafana patch releases on 2026-10-27 fix CVE-2026-100105, CVE-2026-88858 and CVE-2026-88967 across three branches, plus one more in 13.2.
- NIST finalizes NISTIR 8547, its road map for moving off quantum-vulnerable cryptography NIST's final transition report lists which current algorithms get phased out and when, supporting 2030 and 2031 federal migration deadlines.
- CISA adds Node2Shell Brotli flaw CVE-2026-111497 to KEV with a 2026-11-10 federal patch deadline CISA confirms attackers now exploit the Brotli decoder out-of-bounds write behind Node2Shell, days after Microsoft shipped emergency .NET fixes.
- PostgreSQL Anonymizer 3.3 fixes 12 CVEs that let masked roles read real data or gain superuser Dalibo's data masking extension now forbids subqueries in masking rules and SQL-standard function bodies for masked roles to close the leaks.
- Anthropic red team reports AI-built exploit chains for Chrome, Safari, Windows and Linux, all patched An unreleased Anthropic model chained separate flaws into working exploits, including a Linux root shell over multipath TCP fixed 2026-10-15.
- Scan of 7.25 million Gitee repositories finds 26,400 verified-live credentials A researcher ran TruffleHog and Kingfisher over all of Gitee for a month and found live Alibaba Cloud, DeepSeek, Zhipu and OpenAI keys.
- Servicepipe anti-bot script hides its fingerprinting code behind jailbreak prompts and fake bioweapon function names A fingerprinting script from Servicepipe uses bioweapon-themed function names and hidden jailbreak text so AI models refuse to deobfuscate it.
- CISA adds Linux kernel flaw CVE-2026-46300 to KEV after PHP2Shell chain roots 25 firewalls A kernel page-cache write called Fragnesia turned PHP2Shell's nobody shell into root on firewalls, and CISA set a 2026-10-31 deadline.
- Alpine Linux rebuilds its GitLab after exploit attempts on CVE-2026-97229 survived a GitLab upgrade Upgrading GitLab did not fix the CVSS 10 regex flaw on Alpine's instance, because the Ruby patch ships only in GitLab's own packages.
- GMO tells Onamae Desktop Cloud customers to unplug PCs after a breach and an 8 billion yen ransom demand An intrusion may have planted ransomware on up to 26,739 customer servers, and the attacker has started leaking GMO and customer data.
- Attackers probe Oracle E-Business Suite CVE-2026-101767 one week after Oracle's October patch Decoys recorded exploit attempts on an unauthenticated EBS SQL injection that chains to code execution, against about 670 exposed hosts.
- CISA adds Adobe Commerce CVE-2026-48356 and VMware ESX CVE-2026-41709 to KEV A Magento file upload flaw and an ESX logging gap are now known exploited, with federal deadlines of 2026-11-10 and 2026-11-11.
- Skimmer injected into Convertr's self-hosted Plausible script steals cards on Brazilian fashion checkouts Attackers changed an analytics tracker that loads on every Convertr checkout page, then tuned it to the platform's payment fields.
- Microsoft ships out-of-band Windows updates that remove trust in eMudhra roots KB5134133 and KB5134125 drop six eMudhra G3 roots and a cross-sign on Windows machines that do not get automatic trust list updates.
AI
- Anthropic reports results from six drug development partnerships using Claude Claude generated about 35,000 antibody designs for Divergence Labs and helped BigHat raise binding over 10x, per Anthropic's progress report.
- Google DeepMind announces Genie 4 world model with native audio as a trusted tester preview DeepMind's Genie 4 generates real-time interactive worlds with audio and dialogue, steered by Gemini models, starting with trusted testers.
- Liquid AI releases LFM3, six open 2B-class text, vision, audio and omni models on one backbone LFM3-1.7B fits in about 1 GB at 4-bit with a 256K context, and each sibling model fine-tunes on a single GPU for on-device agents.
- NVIDIA releases Nemotron 3.5 Transcribe, an open 30B-A4B model for ASR, diarization and translation NVIDIA releases an open 30B-A4B speech model that transcribes 40 languages, labels up to eight speakers, and ships with the Granary v2 data.