Breaking

Anthropic analyzes 832 misused accounts, warns AI autonomy strains legacy defenses

June 3, 2026 at 14:42 EDT

On June 3, 2026, Anthropic published an analysis mapping the activity of 832 accounts it banned for misusing Claude onto the industry-standard MITRE ATT&CK framework, warning that AI is making attackers more dangerous and that growing autonomy is undermining traditional risk-assessment methods.

June 3, 2026 · Anthropic Threat Analysis

AI Is Erasing the Line Between Low- and High-Skill Attackers

An analysis of 832 banned accounts mapped onto the MITRE ATT&CK framework warns that growing AI autonomy is making attacks more dangerous — and breaking traditional risk assessment.

832

accounts banned & analyzed (Mar 2025 – Mar 2026)

~80%

used the agentic tool Claude Code

100

max ARiES risk score, hit in a Nov 2025 case

Severity is climbing fast

Share of cases rated medium risk or higher — a ~1.7× jump in one year.

First 6 months33%

Latter 6 months56%

The "skill gap" has nearly vanished

Average number of attack techniques used — AI lets low-skill actors chain multi-step attacks too.

Low-skill attackers16 techniques

High-skill attackers~20 techniques

A gap of just ~4 techniques — the "low-skill vs high-skill" distinction is losing relevance.

What AI was used for

Malware creation67.3%

560 accounts — largest category

Lateral movement6.5%

54 accounts — post-intrusion

Shifting later in the kill chain

AI moves from entry points to complex post-compromise work.

−8.6%

Initial-access phishing

+8.9%

Post-compromise account discovery

An inflection point

Autonomous AI orchestration — real-time judgment with minimal human intervention — isn't fully captured by traditional risk frameworks.

MITRE ATT&CK extension in discussion New ARiES risk metric Project Glasswing: 150+ orgs, 15+ countries

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.