From SATAN to Agentic Attacks: We Keep Fearing the Tools, Not the Infrastructure

Here we go again. The cybersecurity community has discovered another terrifying category of tools: agentic attack frameworks. ARTEX, Cairn, PentAGI, HexStrike, and others are making headlines because they allow AI agents to automate reconnaissance, exploitation, and lateral movement. Reports of their use in real intrusions are multiplying, and defensive initiatives such as [Agentic Attack Detections]1 are emerging to track their activities. The concerns are legitimate, but the reaction feels remarkably familiar. We seem much more comfortable discussing the dangers of tools than confronting the uncomfortable reality of the infrastructure they exploit.2

In 1993, Dan Farmer and Wietse Venema published Improving the Security of Your Site by Breaking Into It.3 Their proposal was almost offensively simple: look at your infrastructure the way an attacker would, identify the weaknesses, and fix them. In April 1995, they released SATAN (Security Administrator Tool for Analyzing Networks), making automated vulnerability discovery accessible to a much broader audience. SATAN was not particularly revolutionary in what it discovered. Its real innovation was making existing security knowledge easier to apply. Apparently, that was enough to cause a minor apocalypse in the security community.

The reactions were spectacular. SATAN was accused of putting dangerous capabilities into the wrong hands. On April 3, 1995, CERT published advisory CA-95:06, warning that the availability of an easy-to-use scanner could increase threats against vulnerable systems.4 Dan Farmer lost his job at Silicon Graphics shortly before the release, amid controversy surrounding the tool.5 Even a 1996 US Senate staff statement cited SATAN as an example of increasingly accessible hacking capabilities.6 The recurring argument was already there: making vulnerabilities easier to discover would somehow make the Internet more dangerous. As if the vulnerabilities had been patiently waiting for SATAN before becoming exploitable.

The irony was that SATAN did not create vulnerabilities, and it did not even exploit them. Farmer and Venema explicitly described a tool that identified known problems and provided guidance for fixing them.7 Misconfigured NFS exports, exposed services, outdated software, and broken trust relationships existed long before anyone downloaded SATAN. The underlying problem was the inability, or unwillingness, to maintain secure infrastructure. The scanner simply made that failure visible. Blaming SATAN was rather like blaming the person who switched on the lights in a collapsing building.

Fast-forward thirty-one years. Agentic attack frameworks are undeniably more capable than SATAN. They can chain actions, adapt to failures, explore alternative attack paths, and operate at speeds that make human-driven attacks look almost leisurely. Recent Google Threat Intelligence Group and Mandiant research confirms that these capabilities are being used in real intrusions.8 This is not merely SATAN with an LLM bolted onto it. Yet the underlying targets remain depressingly familiar: vulnerable applications, forgotten services, compromised credentials, excessive privileges, and poorly maintained infrastructure. We have automated the attacker, but have we made comparable progress in maintaining what we defend?

Of course, detection matters. Understanding agentic behavior, building detection rules, and sharing indicators are necessary defensive activities. But we should be careful not to confuse detecting the latest fashionable framework with fixing the conditions that make its success possible. An abandoned server does not care whether it is scanned by SATAN, Nmap, or an autonomous AI agent. An overprivileged account does not become less dangerous because we can identify which framework abused it. And an organization unable to inventory its systems, replace obsolete dependencies, or manage its attack surface will not solve those problems by adding another collection of signatures. We are getting better at recognizing the tools wandering through our infrastructure while remaining remarkably bad at maintaining the infrastructure itself.

Perhaps the most striking parallel with SATAN is how easily our attention shifts toward the technology exposing or exploiting weaknesses rather than the infrastructure that allows those weaknesses to persist. In 1995, we worried about automated vulnerability scanners. In 2026, we worry about autonomous offensive agents. The difference in capability is substantial, but the underlying defensive challenge remains remarkably familiar. We keep debating the tools capable of navigating the ruins while paying too little attention to why so much of our digital infrastructure has been allowed to become ruins in the first place.

  1. ThreatDefence, Agentic Attack Detections, GitHub repository, 2026. An open-source collection of detection rules, hunting queries, and indicators for agentic offensive frameworks. Ironically, we are still largely stacking detection rules in GitHub repositories, hoping defenders will find, maintain, and deploy them. However, projects such as Rulezet offer a more promising direction by automating the discovery, indexing, validation, synchronization, and distribution of detection rules across repositories and security platforms. Perhaps there is still hope: if attackers can automate exploitation, defenders should be able to automate detection engineering too. ↩

  2. Google Threat Intelligence Group, GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, November 16, 2026. ↩

  3. Dan Farmer and Wietse Venema, Improving the Security of Your Site by Breaking Into It, December 1993. The original paper introduced their approach to proactively examining systems from an attacker’s perspective. ↩

  4. CERT Coordination Center, CA-95:06 — Security Administrator Tool for Analyzing Networks (SATAN), April 3, 1995. The advisory explicitly acknowledges that the vulnerabilities identified by SATAN were not new, while warning that an accessible scanner could increase the threat against systems where those vulnerabilities had not been addressed. ↩

  5. Amy Harmon, SATAN Should Deliver a Helluva Birthday, Los Angeles Times, April 5, 1995. See also Rik Farrow, Interview with Dan Farmer, USENIX ;login:, December 2014, for Farmer’s retrospective account of SATAN’s development and the controversy. ↩

  6. United States Senate Permanent Subcommittee on Investigations, Security in Cyberspace, Minority Staff Statement, June 5, 1996. The statement presents SATAN as an example of powerful security tools becoming accessible to individuals with limited technical expertise. ↩

  7. Dan Farmer and Wietse Venema, SATAN — Security Administrator Tool for Analyzing Networks, original release announcement, 1995. The authors explicitly state that SATAN reports common networking security problems without exploiting them. They also acknowledge the dual-use nature of the tool and explain that its purpose is to improve network security. ↩

  8. Google Threat Intelligence Group and Mandiant, ARTEX and Friends: Enter Agentic Offensive Frameworks, October 30, 2026. Documents observed adoption of agentic offensive frameworks in real intrusions, their operational capabilities, and the continuing importance of traditional security controls. ↩