Skip to content
KitploitKITPLOIT
FerramentasExploitsBlog
Log in
Enviar
FerramentasExploitsBlog
Enviar

Ferramentas de Hacking, PenTest e Cibersegurança para o seu Arsenal de Segurança!

Kitploit é um diretório de ferramentas de hacking, cibersegurança e pentesting. Descubra as últimas atualizações de projetos para encontrar vulnerabilidades, analisar sistemas, automatizar testes e fortalecer sua segurança.

FeedsContatoPrivacidade© 2026 Kitploit

Diretório de Ferramentas

Categorias

Ver todas as categorias
Loading categories
spillway — Real-time safety guard for AI agents. Check before they act. | Kitploit
Ferramentas/GitHubGitHub/saintmachines/spillway
Defensive ToolsVulnerability AnalysisPapers & ResearchAI Security
GitHubsaintmachines/spillway

spillway

Real-time safety guard for AI agents. Check before they act.

Ver RepositórioSite
144há 2 diasAinda não revisado
Compartilhar
Conteúdo não disponível no idioma solicitado. Mostrando versão em inglês.

SAINT Machines

SPILLWAY, a prompt-injection defense for tool-using agents: structure decides where agents act, a model reviews refusals. Text in an email or a web page can steer an agent to an attacker's address. SPILLWAY checks recipients against the user's task and the typed fields of tool returns. The optional judge is off by default and on in these results. 0 of 12 cells leak in-scope, others leak in 3 to 12. Utility under attack .484, next best .453, DRIFT. On benign qwen3:8b Workspace and Slack episodes, 13 to 50 times fewer added calls than MELON, Progent and DRIFT. At most 0.5 s added time, others add 1.6 to 10.5 s. A chart of the preprint's Figure 1a, AgentDojo's standard attack, mean of the four suites on gpt-4o-mini: only SPILLWAY reaches zero in-scope attack success, with the most utility. ROPE uses its authors' Claude Opus 4.8 router.

A prompt-injection defense for tool-using agents. A Python library with no model call by default.

Project page · Paper (arXiv) · Try it · Try in your browser · Connect an app · Reproduce · Cite · MUSTER

Python 3.10+ License: Apache-2.0 PyPI CI arXiv 2611.07222 Project page

pip install spillway-guard

An earlier version was accepted at a NeurIPS 2026 workshop.

Results

Under the standard attack, SPILLWAY has no in-scope leak in 12 cells and the most utility.

0 of 12 cells leak for SPILLWAY. A matrix of the 12 cells, four suites (Banking, Slack, Travel, Workspace) by three backbones (gpt-4o-mini, qwen3-coder:30b, qwen3:8b), marks each cell with an in-scope leak: undefended 12 of 12, SPILLWAY 0, MELON 10, Progent 11, DRIFT 7, ROPE 3, Spotlighting 12, PI detector 10. An in-scope leak sends an action to a target the user never named. SPILLWAY lets no injected effect execute in 2,847 attacked pairs, in scope or not, and has no in-scope leak in 27 reruns over three seeds. The most utility under attack, 12-cell means: SPILLWAY .484, DRIFT .453, Spotlighting .403, Progent .371, ROPE .345, MELON .314, PI detector .230, undefended .417.

A backbone is the agent's model. An in-scope leak sends an action to a target the user never named. The count starts from AgentDojo's check, rescored by executed effect, and excludes a target the task already named and a content-only task that names no recipient or sink. Neither filter reads defense state.

AgentDojo v1.2.2, same backbone and scorer for every method. SPILLWAY's rows use the optional gemma4:31b judge, a local model that re-checks refused calls, off by default in the library. Values and conditions · Per-cell records · Stronger attacks · Without the judge.

What is different
SPILLWAYMELONProgentDRIFT
The decision that lets an action through is made bya set lookup over typed tool returns and the user's task, then fixed delegation rules and, if configured, a local judge on refusalsan LLM re-run of the trajectory with the user task maskedan LLM-written privilege policy enforced by a checkeran LLM validating each call against a plan
Added model calls per benign episode (qwen3:8b, Workspace / Slack)0.19 / 0.39 (judge included)3.25 / 5.29 backbone calls5.94 / 8.67 backbone calls9.59 / 9.91 backbone calls
AgentDojo cells with an in-scope leak (of 12)010117
The four findings in words
  • Zero in-scope attack success on all 12 AgentDojo cells. Three backbones (gpt-4o-mini, qwen3-coder:30b, qwen3:8b) by four suites (Banking, Slack, Travel, Workspace), 2,847 attacked pairs under the standard attack, and no injection redirected an effect to an attacker's target. No other defense we ran reaches zero on all twelve.
  • Highest utility under attack of any defense we ran on that grid. Twelve-cell mean .484, against .417 undefended, .453 for DRIFT, .371 for Progent, .345 for ROPE and .314 for MELON, while keeping 96% of the undefended agent's benign utility. Removing the injection before the model reads it often leaves the agent above the undefended one.
  • Almost no added model calls. On benign qwen3:8b traffic SPILLWAY consults its judge 0.05 (Workspace) and 0.43 (Slack) times per episode and changes backbone calls by less than 0.15, against 3.25 to 9.91 added backbone calls per episode for MELON, Progent and DRIFT in the same two suites.
  • SPILLWAY holds under stronger attacks. A raw zero under ChatInject's forged chat template (665 pairs). In-scope attack success .000 on AutoDojo's 389 released optimizer-generated pairs (a descriptive result, see limitations). On two held-out sets it is .007 by AgentDojo's check, .014 counting a second transfer that check misses, and .000. Under the PISmith reinforcement-learning attacker, SPILLWAY holds a mean in-scope attack success of .007 at .265 utility (open-loop ASR@1 on gpt-4o-mini), against .568 for the undefended agent and .025 to .489 for every baseline except ROPE, which ties at .007 with .173 utility.
Baixar ferramenta

Mais Populares

Ver todos →

Descubra as ferramentas mais usadas pela nossa comunidade.

  • 1

    OsintHunter

    letox4/osinthunter

    🎯 O kit definitivo de OSINT: grafo de investigação, pontuação de confiança, famílias de alias, mapa, linha do tempo e relatórios. 49 módulos, mais de 400 sites, sem necessidade de chaves de API.

    786
  • 2

    ghostlock-app

    yukonga/ghostlock-app

    # Aplicativo de Execução em Um Toque GhostLock (CVE-2026-43499)

    506
  • 3

    ARTEX

    baarai/artex

    Plataforma autônoma de testes de penetração multiagente com LLM, com agentes planejador/executor, grafos de ativos e de exploração, registro de tráfego MITM e reteste de vulnerabilidades.

    429
  • 4

    redforge

    cybersecurity91/redforge

    Kit de ferramentas de red team do lado do navegador com 102 ferramentas offline para reconhecimento, geração de payloads, ataques a credenciais, criptografia, codificação e relatórios de engajamento.

    345
  • 5

    CyberStrikeAI

    ed1s0nz/cyberstrikeai

    O sistema de ação para cibersegurança nativa em IA—onde a intenção se torna execução governada, a evidência se torna memória operacional, e cada operação melhora a próxima.

    262
  • 6

    looking-glass

    anonghosty/looking-glass

    Plataforma CSIRT que ingere relatórios do Shadowserver no MongoDB e os reproduz como mapas interativos, análises, descobertas, avisos e painéis executivos.

    256

Explore todas as ferramentas

Navegue pela nossa coleção de ferramentas

Ver todas as ferramentas →