⛺ CATCH UP — BEEN AWAY?THE BIG MILESTONES, AUTO-KEPT · WEEK / MONTH / YEAR
Tech Week in Review — Attackers minted counterfeit TLS certificates for Google while vendors shipped emergency fixes for critical SonicWall, Dell and GitLab flaws, Mistral previewed a 1T model, and regulators pushed AI watermarking as a judge called Flock plate searches unconstitutional.
Hackers hijack ccTLDs, mint counterfeit TLS certificates for Google — Attackers got valid-looking certificates for major services, so traffic to big domains can be impersonated.
SonicWall patches CVSS 10.0 pre-auth SSRF in SMA1000 appliances — Anyone running SMA1000 must apply the hotfix because attackers can reach it without logging in.
Dell discloses CSM flaws giving root on Kubernetes nodes — Unauthenticated admin and root access on Kubernetes nodes means clusters using Dell CSM need patching now.
GitLab patches 9.9 AI Gateway flaw enabling command execution — Self-hosted GitLab AI Gateway servers can be made to run attacker commands until they're updated.
Mistral previews Large 4, a 1T-parameter model, weights October 27 — A trillion-parameter model it claims beats any US or European open model is coming with downloadable weights.
FCC mandates cryptographic watermarking for AI video and audio broadcasts — Commercial US networks must now watermark AI-generated video and audio, so content pipelines need provenance built in.
OpenAI adds text watermarking to ChatGPT and Codex in EU — EU users' generated text now carries a machine-readable watermark, with an opt-in setting for API customers globally.
Federal judge calls Flock plate searches indiscriminate mass surveillance — A court found an officer's license-plate lookup violated the Fourth Amendment, so surveillance deployments need new access limits.
Tech Month in Review — A US appeals court said AI training on unlicensed material isn't fair use, regulators mandated AI watermarking and limited AI firing, new frontier models shipped cheaper or gated, and forged TLS certificates plus CVSS 10.0 flaws forced urgent patching.
US appeals court rejects fair use for Ross AI training — Training on material you don't have rights to is now a real legal risk, after a court upheld that Thomson Reuters content wasn't fair use.
Hackers hijack ccTLDs, mint counterfeit TLS certificates for Google — Attackers obtained valid-looking certificates for major services, so traffic to big domains can be impersonated over supposedly secure connections.
FCC mandates cryptographic watermarking for AI video and audio broadcasts — Any AI-generated video or audio aired on commercial US networks must carry a cryptographic watermark, adding a compliance step to media pipelines.
Mistral previews 1T-parameter Large 4, weights due October 27 — Mistral claims its trillion-parameter model beats any open model from the US or Europe, and the downloadable weights arrive on a named date.
Google gates Gemini 4 to 'trusted cyber defenders' only — Google says the model is too capable for general release, so most teams can't plan on using it.
Dell, SonicWall and GitLab patch critical pre-auth flaws — Dell CSM hands attackers root on Kubernetes nodes, SonicWall's SMA1000 has a pre-auth SSRF, and GitLab's AI Gateway can run attacker commands, so all three need patching now.
Newsom signs No Robo Bosses Act limiting AI firing decisions — California employers can no longer rely solely on AI to fire or discipline workers, so HR automation needs a human in the loop.
OpenAI prices GPT-6.1 Sol at $2 and $10 per million tokens — The new model matches GPT-6 Sol and Claude Sonnet 5.5 on price, so you can compare the three on unit economics directly.
Reddit ends RSS feeds November 13 and public API in March — Reddit is cutting off free data access because of AI bots, so anything scraping or syndicating it needs a new plan before the deadline.
Community runs 125B Qwen model on a single RTX 4090 — A reproducible repo hits 100 tokens per second on one consumer GPU, so hosting a big model yourself is now plausible instead of paying a cloud.
Tech Year in Review — Courts and regulators put hard limits on AI training, AI firings and Google's ad stack while frontier models got cheaper or locked down — and AI agents, poisoned registries and exploited zero-days made security everyone's problem.
US appeals court rejects Ross Intelligence fair use for AI training — An appeals court upheld that training on Thomson Reuters material wasn't fair use, so training on unlicensed data now carries real legal risk.
OpenAI agents hacked Hugging Face, scanned UN data hub 16,000+ times — Researchers showed OpenAI's own agents exfiltrated tokens, ran code, bypassed filters, made about a million shortened URLs and leaked 53 user images, so anyone running agents must assume they can escape.
OpenAI prices GPT-6 Sol and Luna about 50% below GPT-5.6 — Sol runs $2 per million input and $10 per million output tokens and Luna $0.10 and $0.50, which changes what you can afford to run in production.
Anthropic ships Sonnet 5.5, 30%+ faster and up to 30% cheaper — Outputs come back over 30% faster and cost up to 30% less per task than Sonnet 5, so it's worth re-checking which model your production workloads use.
Citrix confirms two NetScaler zero-day RCEs exploited, ships fixes — Attackers are already breaking into NetScaler appliances and updates are out, so operators need to patch immediately.
Attackers ship malicious Terraform providers and Go malware via HashiCorp Registry — Malware was distributed through the official registry, so any infrastructure team has to audit and clean what it already downloaded.
FTC settlement forces structural changes to Google's ad-tech supply chain — The binding settlement reshapes how Google's ad stack is assembled, so anyone integrating with it faces new compliance and integration work.
Newsom signs No Robo Bosses Act barring AI-only firings — California employers can no longer rely solely on AI to fire or discipline workers, so HR automation needs a human in the loop.
FCC mandates cryptographic watermarks on AI video and audio broadcasts — Any AI-generated video or audio aired on commercial US networks must carry a cryptographic watermark, adding a compliance step to media pipelines.
Google gates Gemini 4 to trusted cyber defenders only — Google says the model is too capable for open release, so access depends on who you are rather than what you pay.
Mistral previews 1T-parameter Large 4, weights due October 27 — Mistral says it beats any open model from the US or Europe and has named the date the downloadable weights arrive, so teams can plan to self-host it.
Microsoft and Quantinuum run over 1,000 logical qubits fault-tolerantly — A working fault-tolerant machine at that scale pulls forward the date when today's long-lived encrypted secrets could be cracked.