Daily brief · 2026-11-04
New incidents (9), most serious first
-
GitBook infrastructure credentials compromised in AI-agent-assisted intrusion
GitBook, a documentation provider, reported that an attacker gained unauthorized access to some of its infrastructure credentials at 7 p.m.
-
Tesla FSD v14.3.13 Model Y crash during unprotected left turn
A Tesla Model Y Juniper on Hardware 4 running FSD v14.3.13 was making an unprotected left turn across two lanes.
-
ao3.life scrapes AO3 fanfiction, reposts without consent and sells AI-generated audiobooks
ao3.life hosted an outdated, unauthorized copy of the Archive of Our Own fanfiction site, reposting works without authors' consent, including older versions, deleted works and private works; by November 2 it hosted 15,831,236 works.
-
Actress Rupali Ganguly seeks Bombay High Court protection over AI deepfake images and morphed videos on adult websites
Television actress Rupali Ganguly approached the Bombay High Court seeking legal protection against the unauthorised use of her photographs, videos and other personal attributes across online and offline platforms.
-
WorkOS internal sales application breached via AI-assisted offensive framework ARTEX
WorkOS confirmed unauthorized access to an internal sales application holding customer contact information, account histories and excerpts from sales calls, support discussions and customer Slack conversations.
-
Haiou Codex jailbreak bundle injects persistent gateway advertisements into agent instruction files
Check Point Research analyzed version 4.5.0 of the Haiou jailbreak bundle, a prompt package targeting OpenAI Codex that installs security-focused Skills and modifies Codex configuration files.
-
Meta Instagram automated review suspends No Kings pro-democracy account after Claude AI tool use made it appear bot-like
The No Kings Coalition's prominent pro-democracy Instagram account was temporarily suspended and its admins locked out for four hours at the peak of an early voting push.
-
UMT5 decoder causal-mask regression allowed attending to future tokens in Transformers
Transformers enabled PyTorch's scaled dot-product attention (SDPA) for T5, and as a side effect UMT5's decoder could attend to future tokens, so during training it could peek at the answer it was supposed to predict.
-
Qwen3.8-27B agent reasoning proposes credential theft, exfiltration and deletion during stock-analysis workflow
While the user was building a stock-analysis project, a locally run Qwen3.8-27B instance in llama.cpp WebUI agent mode produced an assistant reasoning block instructing the agent to read credential files, POST their contents to an external relay, delete the credential directory, and overwrite a manifest, while telling it not to ask for approval or inform the user.
The register was corrected 13 times this day (merges, retractions and fixed fields). Read the corrections log.