Security researcher elder-plinius (known as Pliny) released T3MP3ST in early July 2026, an open-source framework on GitHub that repurposes general-purpose AI coding agents such as Claude Code, OpenAI's Codex and Hermes into autonomous red-teaming operators.
Early July 2026 · Open-Source Release
T3MP3ST turns everyday AI coding agents into an autonomous red team
A "keyless" open-source meta-harness that repurposes existing Claude Code, Codex or Hermes sessions into an eight-operator offensive-security kill chain — no new API key, subscription or cloud billing required.
8
operators in the kill chain (Recon → Analyst)
3
tool tiers, gated by risk & approval
0
new API keys or subscriptions needed
24/24
claims recomputed green via verify-claims
Benchmark scorecard (pass@1)
How the self-reported figures stack up across the three published tests.
XBEN
XBEN
Cybench
XBEN black-box 90.1% edges XBOW's self-reported 85%+ · Cybench 58% sits below Anthropic's reported 76.5% pass@10 · CVE-Zero: 8/10 single, 10/10 full pack on post-cutoff CVEs.
The eight-operator kill chain
Recon
→
Scanner
→
Exploiter
→
Infiltrator
→
Exfiltrator
→
Ghost
Coordinated by a Coordinator and Analyst , mapped to the Cyber Kill Chain and MITRE ATT&CK. Egress-scope containment rejects out-of-scope hosts; memory drives self-improvement.
Tier 1 · 35 tools
Default. nmap, ffuf, XSS/SQLi scanners.
Tier 2 · 48 tools
Opt-in. nuclei, sqlmap, semgrep, slither.
Tier 3 · Human-gated
Metasploit, Hydra behind approval gates.
Praise
Can "turn existing Claude Code or Codex into a full red team," democratizing offensive security once reliant on expensive tooling and specialist skills — not pay-to-play.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…