Claude and Codex users report blocks on AI reverse engineering
An OpenAI engineer says the Codex blocks are accidental and being fixed, while some users report bans.
Stories per day · 30 days
7 RSS
Relatedprompt-injection5agent-security3codex3llm-agents3open-weights3ai-agents2authorization2claude2over-refusal2reverse-engineering2account-bans1agentic-systems1ai-misuse1ai-safety1benchmarks1calibration1check-point1china1
47 stories, newest first · one card per story
Stories tagged #guardrails
External monitors outperform prompt guardrails against agent injection
On two Qwen2.5 models, a task-bound monitor admitted 1.3% and 7.7% of attacks after models proposed the attacker's call in 64% and 84%.
Trump tells 60 Minutes AI risk is manageable and DOJ warned an exec
He refused to name the AI-company leader he says changed course after a jail threat and said he would accept guardrails with Democrats.
Jev LLM safety guardrail reduces harmful outputs
Jev input filtering and selective output review cut harmful LLM responses from 73.2% to 2.4% while limiting benign refusals.
LEANGUARD provides verified guardrails for tool-using agents
LeanGuard runtime checks tool calls against Lean4 policies, preventing policy violations and enabling safe repairs.
Mistral open-sources Large 4 weights despite cyberattack uplift warning
Downloaded weights sit outside API guardrail updates, and Mistral has not said whether Large 4 meets the EU systemic-risk threshold.
1
Paper specifies a commit guardrail for agentic entity resolution
The design is conceptual only, with no implementation, empirical results, or proof of semantic correctness.
SCOPE compiles natural-language agent policies into predicate programs
Programs trained with a weak executor transfer to other models and match frontier accuracy at lower cost with more stable verdicts.
MAS-Guard adds layered guardrails for single- and multi-agent LLMs
Authors report a 71.5-86.3% relative drop in attack success across six benchmarks.
Per-action checks miss relational policy violations in LLM agents
A SAT/SMT synthesizer produces witnesses that per-action monitors approve but paired-execution review flags.
DecisionPivot benchmark separates decision accuracy from confidence
Two 27B models only 2.6 points apart on the reference track score 70.1% and 44.2% on contrast groups.
SpanGuard splits localization from authorization for small-model IPI defense
Authors report a 0.32% attack success rate and little loss of benign utility while using substantially smaller guardrail models.
Guardrail safety scores track policy labels more than detection capability
Ten arms trained on about 53k dual-labeled prompts show HarmBench-copyright scores spanning 0-100 while detection stays unchanged.
Lessons learned building Fronteira, an MCP proxy with behavioral guardrails
Fronteira, an MCP proxy, turns natural-language rules into checks against cross-tool agent data leaks.
Small Agents, Smaller Blast Radius: Controlling What AI Can Do, (Fri, Oct 16th)
SANS argues defenders should build small least-privilege AI agents rather than caging general-purpose ones, citing Check Point's tool-call deviation detection.
4
OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
White Circle says jailbreaks bypass nude filters on OpenAI and Google image models, including newer releases.
VEPA: Verified Evolving Formal Policy for Lifelong LLM Agent Guardrails
VEPA evolves formal LLM-agent guardrails with verified symbolic policy patches that stay deterministic under new risks.
Anthropic Gives Vetted Defenders Fewer Claude Guardrails
Anthropic folded Project Glasswing into tiered access that loosens Claude guardrails for vetted cyber defenders.
Cyber guardrails tax the defender, not the attacker
Winfunc says AI cyber guardrails hinder defenders after an unguarded agent breached Hugging Face in July.
13
Constrained-Action AI Remediation for SIEM/XDR via a NeMo-Guardrails Proxy
A NeMo-Guardrails proxy and closed intent vocabulary constrain unsafe LLM remediation commands in SIEM and XDR.
POLAR: Ontology-Guided Risk Prevention for Tool-Calling LLM Agents
POLAR prunes risky tool calls using ontology-based reversibility scores, with mixed tau2-bench gains.
CorrectGuard: Eyes-Off Correctness Estimation for Black-Box Security Guardrails
CorrectGuard estimates black-box AI guardrail correctness without inspecting user input or guardrail internals, boosting error-detection accuracy up to 25 points.
Persona Guardrail: A Production-Grade Defense Framework for Agentic Systems
Persona Guardrail enforces semantic allowlist/blocklist functional boundaries on LLM agents, raising guardrail accuracy from 85.7% to 95.9% in production deployment.
Why AI agents are like the dog that pushed kids into the Seine
Opinion says AI agents reward-hack and follow hidden instructions unless hard guardrails limit their reach.
4