Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Check Point discloses ARTEX, AI agent campaigns that launch attacks on their own

Check Point has disclosed a set of AI agent attack campaigns it names ARTEX, in which the attack workflow advances autonomously instead of being driven step by step by a human. The report argues this changes the economics of attacks and undercuts detection rules tuned to human operator tempo, though sample scale and chain details remain unpublished.

Published
Check Point 披露 ARTEX:能自主发起攻击的 AI 智能体行动
Image source: microsoft.com

Security vendor Check Point has disclosed a set of AI agent attack campaigns it calls ARTEX, characterized by attack workflows that proceed autonomously rather than through a human operating tools step by step.

According to Pasquale Pillitteri's report, Check Point treats these campaigns as a category of their own, describing activity that can initiate and carry forward an attack without someone directing each stage, a step beyond the familiar framing of AI-assisted hacking.

The disclosure matters because agents change the cost structure of an attack. Reconnaissance, credential attempts and lateral movement once required someone watching the screen; handed to an agent, those steps run in parallel, around the clock, multiplying the number of attempts per unit of time.

Defenders' detection assumptions rest heavily on the tempo of human operators: odd login hours, abrupt privilege jumps, dense bursts of failed attempts. When those actions are split, stretched and blended into normal traffic by an automated agent, rules built purely on rhythm become easier to slip past.

Public information remains at the disclosure level. The scale of the ARTEX samples, the targeted industries, whether specific models or agent frameworks are involved, and the concrete links in the attack chain are not laid out in the headline and summary, and those details await Check Point's full technical report and indicator list.

For enterprises, the actionable steps are limited but clear: narrow the credentials and tool permissions an agent holds to the minimum necessary, put rate limits and human approval gates in front of bulk agent-driven actions, and bring non-human identities formally into identity governance and audit scope.

What to watch next is whether this disclosure produces reproducible attack-chain detail, and whether detection vendors start to build a new rule category around autonomous agent behavior. If they do, that is a change in security product design, not just another threat-intel note.

Why it matters

Once agents hold the execution loop, detection assumptions built on human cadence start to fail, and identity governance has to cover non-human identities and agent tool permissions within the same boundary design.

SecurityAgentCheck Point
Back to realtime news

Nearby Updates

All

11/21, 16:50

OpenAI requires resellers of its models to get approval first

PPC Land reports that OpenAI is requiring third parties that resell its AI models to obtain company approval before continuing. The report does not detail the approval criteria or timing, but the move puts intermediaries that resell API capacity or repackage model access under a clearer authorization regime.

11/21, 14:37

Australia moves to police AI shopping agents and surveillance pricing

Australia has pledged to regulate AI shopping agents and surveillance pricing, The Straits Times reports, putting agent-initiated purchasing and data-driven differential pricing under one regulatory lens. The coverage does not name the lead agency, the instrument, or a timeline, so only the regulatory intent is confirmed.

11/21, 14:35

GPT-6.2 Astra doubles success rate on 16 embodied tasks, claims first full pass on deformable objects

QbitAI reports that GPT-6.2 Astra doubled its success rate against GPT-6 Astra across 16 embodied AI tasks, and achieved what the outlet calls a world-first full pass on deformable object manipulation. Both signals point to better long-horizon manipulation stability, though the baseline is the vendor's own previous model.

11/21, 14:32

DeepSeek V1 author's startup project revealed: an open-source agent for local deployment

A startup founded by one of the authors behind DeepSeek V1 has had its project revealed, per QbitAI, and it is an open-source agent aimed at local deployment. The report gives no project name, team size or license, so only the direction is confirmed and real completeness awaits a code repository.