AIQB
TutorialsOrdinary

The gate that stops an agent's memory from re-firing an attack

Source: DEV Community·

Summary

The carryover problem Most prompt-injection work treats the attack as a one-turn event....
CategoryAI Tutorials & Practice
TierOrdinary
Published
Indexed by AIQB
SourceDEV Community
AIQB record IDintel-f37ccd695e8905a99563fe37
Read original source ↗More Tutorials

Related stories

Matched by category, entity tags and source

TutorialsOrdinary
The Judge Reads Tokens: An Adversarial Evaluation of an Eval Toolkit's Own Scorers
The Judge Reads Tokens: An Adversarial Evaluation of an Eval Toolkit's Own...
Source: DEV Community
TutorialsOrdinary
The Test Passed. The Coupon Was Still Wrong.
Building SYNKAI with Switch: a coding agent, an independent verification boundary, and a human...
Source: DEV Community
TutorialsOrdinary
How I audit AI-written code — and the bug that proved why
I ship as an AI-augmented solo builder, but I never trust what the model produces. Here's the review discipline that caught a missing Android permission in every earlier release build.
Source: DEV Community
TutorialsOrdinary
Uneven Branches Make a LangGraph.js Join Node Run Twice. Super-Steps Explain Why
One branch has one step, the other has two, both feed combine. LangGraph.js runs combine twice. Why, and the two fixes: array edges and defer: true.
Source: DEV Community
TutorialsOrdinary
A human was about to approve the payment. Then your context compaction ran.
Automatic context compaction and human approval gates are on a collision course. A one-line entry in the OpenAI Agents SDK changelog names the bug, and the fix is an approval record that lives outside the conversation.
Source: DEV Community · OpenAI
TutorialsOrdinary
499 requests called themselves AI bots on our site in 14 days. We could verify 105.
Six AI bot names in 14 days of logs on one site: 105 of 499 requests verified, 176 from outside the operator's IP list. How to check your own.
Source: DEV Community
TutorialsOrdinary
Your LLM Agent Is an Untrusted Employee: 6 Security Guardrails We Built for AI in Banking Operations
Prompt injection, hallucinated numbers, runaway costs and leaked card data: practical guardrails for putting AI agents into a security-sensitive platform.
Source: DEV Community
TutorialsOrdinary
A client's vibe-coded app was leaking data. Here's how I closed it in 24 hours
A client's internal app (an inventory dashboard connected to 4 live Shopify stores) had been built...
Source: DEV Community
TutorialsOrdinary
The coding-agent toolbox around the model: 8 notes on CLAUDE.md, mcp2cli, OpenCode and worktrees
Once you use a coding agent every day, the model itself stops being the main lever. What changes the...
Source: DEV Community · Claude
TutorialsOrdinary
It Remembers Across Restarts — and Forgets Mid-Conversation
Session files stored fourteen messages while the model saw six. A restart keeps your name, and seven turns later the same conversation forgets it.
Source: DEV Community