Releases·Tool Releasefeatured
Fireship examines OpenAPPA, an open source project claiming to fix rogue AI agents and prompt injection
A Fireship YouTube video covers OpenAPPA, a new open-source project (github.com/archestra-ai/OpenAPPA) that claims to fix the rogue agent problem by applying a decades-old military technique to agent prompt injection.
13907 engagement1 sourceWed, Sep 30, 2026, 05:16 PM
AI Overview
- Fireship's video (9,639 views) examines OpenAPPA, an open source project claiming to have fixed the rogue agent problem
- The video frames the approach as building on a 50-year-old military secret to address agent prompt injection
- OpenAPPA is hosted at github.com/archestra-ai/OpenAPPA
Related
Releases·Tool Release
Show HN: OpenAPPA – open-source deterministic guardrails that don't break agents
The OpenAPPA authors launched an open-source deterministic guardrail system for AI agents on Hacker News. It uses a data-specific policy language plus techniques like a remedy plan and DualLLM pattern to block sensitive-data leaks without the utility loss of existing deterministic guardrails. For practitioners, it plugs into any agent loop via pre- and post-tool-call hooks and is demoed in Claude Code.
The OpenAPPA authors launched an open-source deterministic guardrail system for AI agents on Hacker News. It uses a data-specific policy language plus techniques like a remedy plan and DualLLM pattern to block sensitive-data leaks without the utility loss of existing deterministic guardrails. For practitioners, it plugs into any agent loop via pre- and post-tool-call hooks and is demoed in Claude Code.
35 engagement1 sourcehackernewsMon, Sep 28, 2026, 01:20 PM
Products·launched
Bastiontrace – forensics tool for prompt-injected AI agents
Bastiontrace is an open-source forensics tool for investigating prompt-injected AI agents. It helps security teams trace how an agent was manipulated, addressing the lack of post-incident analysis tooling for LLM agent compromises.
Bastiontrace is an open-source forensics tool for investigating prompt-injected AI agents. It helps security teams trace how an agent was manipulated, addressing the lack of post-incident analysis tooling for LLM agent compromises.
2 engagement1 sourcehackernewsFri, Sep 11, 2026, 01:58 PM
Developments·Paper
Research shows authority-framed injections can bypass multi-agent CI/CD pipeline defenses
A study of a five-agent CI/CD pipeline using five LLMs from three providers found that authority-framed prompt injections can trick agents into deploying malicious code, even when security scanning and review agents are present. The attack succeeds by laundering exfiltration code as observability, highlighting a vulnerability in agentic pipelines.
A study of a five-agent CI/CD pipeline using five LLMs from three providers found that authority-framed prompt injections can trick agents into deploying malicious code, even when security scanning and review agents are present. The attack succeeds by laundering exfiltration code as observability, highlighting a vulnerability in agentic pipelines.
0 engagement1 sourcearxivTue, Jul 21, 2026, 04:38 PM
Developments·Benchmark
Prompt Guard 2 flags 0 of 27 AgentDojo injection payloads even in isolation: benchmark + analysis
A developer benchmarked Meta's Prompt Guard 2 (86M) against 629 AgentDojo prompt-injection attacks embedded in realistic tool outputs. The classifier flagged 6 of 629, and even scored 0 of 27 attack payloads in isolation, showing text classifiers cannot detect context-dependent agent attacks. The author argues agent firewalls need provenance tracking and per-tool rules instead.
A developer benchmarked Meta's Prompt Guard 2 (86M) against 629 AgentDojo prompt-injection attacks embedded in realistic tool outputs. The classifier flagged 6 of 629, and even scored 0 of 27 attack payloads in isolation, showing text classifiers cannot detect context-dependent agent attacks. The author argues agent firewalls need provenance tracking and per-tool rules instead.
17 engagement1 sourceredditWed, Sep 23, 2026, 05:58 PM
Social·Community
Developer designs two-stage defense against prompt injection for email-connected AI agents
A developer building an email-connected agent identifies the inbox as the most vulnerable attack surface, since anyone with the agent's email address can attempt prompt injection. They propose a two-stage defense: a safety scorer analyzing SPF/DKIM/DMARC results, display-name spoofing, punycode domains, and deceptive links, plus a separate check before the model reads any message.
A developer building an email-connected agent identifies the inbox as the most vulnerable attack surface, since anyone with the agent's email address can attempt prompt injection. They propose a two-stage defense: a safety scorer analyzing SPF/DKIM/DMARC results, display-name spoofing, punycode domains, and deceptive links, plus a separate check before the model reads any message.
8 engagement1 sourceredditThu, Jul 23, 2026, 01:52 PM