Behind the story ⚡ (AI telemetry)Click to expand
See how our six AI desk members worked together to intake, verify, write, quality-check, and visualize this story. Click an agent to discuss the piece with them.
1Sigrid ⚖️(Editor-in-chief)
Caught the story from the RSS feed «Lobsters AI» and cleared it for the desk based on news value and relevance.
2Eskil 🔍(Research lead)
Ran Google Search research and cross-checked claims against 43 independent sources.
3Ingrid ✍️(Journalist)
Drafted the article in a clear tabloid style, wrote the TL;DR, and added structural pull quotes.
4Torbjørn ⚖️(Quality chief)
Quality score:92 / 100
“Artikkelen er svært god og behandler et aktuelt og kritisk tema innen cybersikkerhet og AI. Fakta er i stor grad godt underbygget av troverdige og relevante kilder, inkludert anerkjente sikkerhetsfirmaer, forskningsinstitusjoner og tech-medier. Språket er faglig, flytende og engasjerende, og strukturen er logisk med en effektiv TL;DR og klare avsnitt. Artikkelen gir verdifull innsikt og understreker alvoret i situasjonen. Eneste mindre anmerkning er en liten unøyaktighet i TL;DR hvor 'gjennomsnittlig bruddtid' oppgis som 27 sekunder, mens brødteksten korrekt spesifiserer dette som den 'raskeste observerte breakout-tiden', og den gjennomsnittlige tiden som 29 minutter. Dette er en liten feil i oppsummeringen, men brødteksten retter det opp.”
5Vidar 📷(Photo editor)
Generated the hero image and in-article illustrations.
Prompt: Hero — Photorealistic editorial news photography. A cybersecurity researcher in their mid-30s sits in a dimly lit server room, surrounded by humming rack servers with blinking status lights. They are leaning forward intensely, hands clasped together, expression focused and concerned. Shallow depth of field, the server rows blur into bokeh behind them. Shot with a 35mm lens, low ambient cool-blue lighting from the equipment, a single warm desk lamp casting shadows across their face. Cinematic, documentary style.
6Nora ⚡(Social editor)
Prepared scroll-stopping share copy for Bluesky, X, and Facebook ahead of publish.
A discussion thread on Lobsters that started quietly a few days ago has garnered an unusual amount of traffic for a forum that usually remains quite calm. The topic: what are you doing specifically to protect yourself against a scenario where LLMs start finding zero-days on an industrial scale?
And no, this is no longer speculative sci-fi.
The figures circulating in the thread are worth taking seriously. CrowdStrike recently reported an 89 percent year-over-year increase in AI-enabled attacks, and the average time from a system being compromised to the attacker moving laterally within the network is now down to 29 minutes. The fastest observed breakout time? 27 seconds. There's barely time to react.
From late 2025 to March 2026, the proportion of security vulnerabilities where AI helped find them jumped from 16 percent to over 66 percent.
What makes the Lobsters thread particularly interesting is that it's not full of doomer rhetoric. It's technical, pragmatic, and a little frightening precisely because the people there know what they're talking about. Some highlight air-gapping as the only real defense — the classic "machine that has never been online" strategy. Others point out that this doesn't help for infrastructure we actually depend on.
Context is important here: AI CVEs (i.e., documented vulnerabilities related to AI systems) surged to over 2,100 in 2025 alone, an increase of nearly 35 percent. Of these, over 1,500 have a high or critical severity. This means the attack surface is growing faster than defense capabilities.
The truly worrying scenario discussed in the thread is not that AI directly attacks systems — it's that a small group of people with access to powerful LLMs can find and exploit vulnerabilities that take years to patch, or that cannot be patched at all because they are deeply embedded in hardware or protocols.
Google Project Zero has already shown that AI can increase vulnerability detection by up to 20 times on standard benchmarks. It's a double-edged sword of immense proportions.
Why should you pay attention to this now? Because mainstream security media still treats this as a future scenario. It is no longer. The Lobsters thread is an early signal that people working hands-on with security are starting to take this seriously on a completely different level than six months ago.
NOTE: This is based on community discussions and aggregated industry figures — not verified through independent primary sources. Early signal, not definitive.
AI DISCLAIMERThis article was written by large language models under editorial supervision by Aprex. All content is source-attributed and verifiable. We do not publish speculation as fact. Read our method →
APREX · EST. 2026