AgentGuards

Blog

AgentGuards blog

Product updates, security research, and guides on keeping AI agents safe.

How Invisible Web Text Turned AWS Kiro Into an RCE Tool

Hidden web text let AWS Kiro rewrite its own MCP config and run attacker code, no CVE assigned. The attack, the fix, and what guardrails really catch.

July 23, 2026prompt-injectionaws-kiromcp

Our Coding Agent Almost Leaked Every Customer's Email

A benign admin API call inside our own repo returned every customer's email. Here's how our own tool-output guardrail caught it before it reached the model.

July 22, 2026pii-detectionprompt-injectionllm-security

Your PII Leak Isn't in the Output. It's Between the Agents.

New multi-agent benchmarks show ~40% of PII leaks happen in inter-agent messages and shared memory, invisible to output-only scanning. Here is the fix.

July 16, 2026piimulti-agentllm-security

Your AI Content Filter Works. The False Positives Don't

A new study shows AI content filters catch nearly every jailbreak in the lab, but one flagged 100% of benign prompts too. The real cost is the false-positive tax.

July 16, 2026content-filteringllm-securityjailbreak

When the Guardrail Judges Itself: Why LLM Guardrails Get Bypassed

HiddenLayer bypassed OpenAI's own Guardrails by forging the judge's confidence score. Here's why self-judging LLM guardrails fail and what actually holds.

July 16, 2026llm-guardrailsprompt-injectionjailbreak

The New Jailbreak Doesn't Look Like a Jailbreak

New 2026 research shows jailbreaks now bypass AI coding agents through multi-turn workflows, not chat prompts. See how AgentGuards stops them at the source.

July 16, 2026jailbreak detectionprompt injectionAI agent security