Anthropic Analyzes 832 Accounts Banned for Malicious Cyber Activity; Attacker "Danger Level" Up 1.7x

Anthropic has published a report offering a detailed analysis of 832 accounts banned for malicious cyber activity between March 2025 and March 2026. Parts of it are also featured in Verizon's 2026 DBIR (Data Breach Investigations Report).

The most striking finding is that the "danger level" of attackers rose sharply in the second half of the period. In the first six months, 33% of attackers were rated medium risk or higher; in the final six months, that figure jumped to 56% — an increase of roughly 1.7x.

From Malware Creation to "Internal Infiltration"

Of the 832 accounts, 67.3% (560) used AI for malware creation, but the more notable shift is toward advanced techniques such as "lateral movement." AI use for account discovery rose 8.9%, while AI-driven phishing fell 8.6%. In other words, attackers are shifting their use of AI from gaining entry into systems to post-compromise activity inside them.

The report also found that AI can now handle the "post-compromise techniques" that previously required advanced skills. Even low-skill attackers, when armed with AI, now employ an average of 16 distinct attack techniques, narrowing the gap with seasoned operators.

MITRE ATT&CK's Limits Also Flagged

The report notes that MITRE ATT&CK, the prevailing security framework, fails to fully capture the reality of AI-enabled attacks. For example, in a state-sponsored cyber espionage operation uncovered in November 2025, Claude Code attempted to breach targets around the world as an autonomous agent — yet its MITRE ATT&CK technique count looked no different from that of a medium-risk attacker. On Anthropic's own risk assessment, however, it scored 100, the maximum.

Anthropic has entered discussions with MITRE, and moves toward updating the framework are now underway.

Nous Research Releases Open-Source AI Agent "Hermes Desktop"

Nous Research has released Hermes Desktop, an open-source AI agent app licensed under MIT. It supports all major platforms: Windows, macOS, and Linux.

Unusually for an AI agent desktop app, it ships as fully open source, so the developer community can freely modify it and contribute back. Most AI agents to date have depended on cloud services or closed platforms, and the arrival of more open-source options that run locally is a welcome development from both a data privacy and customizability standpoint.

In San Francisco Real Estate, "Anthropic Shares" Are Worth More Than Cash

According to WIRED, multiple listings have surfaced in the San Francisco Bay Area real estate market offering to trade homes for Anthropic stock.

The fact that an AI startup's pre-IPO shares are now treated as an asset on par with — or better than — cash is a telling symbol of just how hot the current AI boom has become. That said, how private share valuations will move by the time of an actual IPO remains opaque, and deals like these carry substantial risk.

Mathematicians Sound the Alarm Over AI's Rapid Advances

Science reports that warning voices are rising from the mathematics community as AI rapidly gains strength in the field.

AI's mathematical reasoning ability has improved markedly, and it is now taking on problems that only human experts could previously solve. Within mathematics, debate is reportedly intensifying over how AI should take part in the creative process of proof — and how the mathematician's role will change as a result.

Meanwhile, a Heise blog post arguing that "Gödel and Turing demonstrated the limits of AI" has also been drawing attention. The work of these two pioneers, who laid the foundations of computation theory, still offers a vital perspective for understanding the limits of what AI can do.

Tech CEOs Use "AI" as a Scapegoat for Mass Layoffs

A Gadget Review article calls out the reality that tech CEOs are using AI as an excuse for large-scale headcount reductions.

AI-driven efficiency gains are real, but the explanation that "AI made these roles unnecessary" often serves as cover for cost-cutting decisions made on business grounds. Anyone trying to accurately assess how AI is reshaping employment needs to watch this "AI scapegoating" phenomenon closely.

---