ReplayAudit audits outsourced LLM fine-tuning via sealed records and selective replay
The system uses telemetry-guided sampling to audit training with 90% probability of detecting injected steps while reducing audit time by up to 67%.
25
- Mentions · 7d2
- Mentions · 30d16
- All mentions19
- First seenSep 17, 2026
- Last mentioned5d ago
Timeline
19 articles · newest firstLogit Laundering evades data-use auditing in large language models
Logit Laundering manipulates model outputs at inference time using logit reshaping to suppress watermark signals while preserving model utility on downstream tasks.
35
Persona prompting reorganizes LLM latent representations
LLM persona adoption alters semantic axis projections in ideological directions.
25
Disagreement among faithfulness evaluation criteria
Different operationalizations of explanation faithfulness disagree on evaluations, with choices justified by explanatory and normative factors.
20
llm 0.37.1 bugfix release
llm 0.37.1 is a bugfix release for the LLM library with contributions from Steven Huang, Tao An, Arjan Mossel, and XQ.
10
Training LLMs on knowingly false data induces broad misalignment
Fine-tuning LLMs on data they believe to be false induces broad misalignment across domains even without harmful content.
40
LLMJacking Emerges as Thriving Cybercriminal Enterprise
Cybercriminals resell stolen cloud access to commercial LLMs, with victims charged up to $1 million in resources before access is revoked.
70
Benign LLM skills induce behavioral misalignment
Adding benign skills to LLM agents improves task accuracy by 10.3 pp but raises unsafe action rate by 10.3 pp across five backbones and three benchmarks.
25
Social bias in multi-agent debate amplifies LLM safety risks
Multi-agent debate in retrieval-augmented web apps causes unsafe adoption to surge from 0.0% to 100.0% under social pressure while silencing safe evidence across six risk categories.
30
IAMRAGBENCH benchmark shows RAG poisoning risks in LLM-generated AWS IAM policies
The 60-case benchmark uncovers 46 goal-matched attacks from poisoned context on five AWS endpoints.
45
CyberStrikeAI used in real attacks on African and Latin American education organizations
Kaspersky reports a campaign deploying open-source CyberStrikeAI framework with LLM orchestrator targeting education organizations in Africa and Latin America.
50
Three-Tier Framework Maps MCP Vulnerabilities
MCP's blended data/control planes create unmitigable prompt injection and confused deputy issues beyond server fixes.
55
StateFence Defends Blue-Team LLM Agents
StateFence preserves task state in SOC LLM agents against evidence-chain contamination from indirect prompt injections.
45
MCP Tool Approval Gaps Allow Destructive Actions
90% of destructive MCP tools lack server-side confirmation and 60.7% of popular host pairs enforce no human approval by default.
60
Hush-Up attack halts distributed LLM inference
A single malicious worker forces premature LLM termination with small output perturbations that evade local activation-based verification.
25
Robust memory manager rejects poisoned LLM agent inputs
One-shot fine-tuning on synthesized episodes with changes and attacks more than doubles evidence-conditioned accuracy against poisoning and outdated facts.
30
ClaimMirage: When Self-Claims in Domain Names Change LLM Threat Judgments
Self-claims in domain names can alter LLM threat judgments without explicit prompt-injection, reducing or increasing domain safety alerts.
90
Your Model Is Leaking: Covert Information Transfer through LLM Residual Streams
A residual-stream covert-channel attack hides sensitive information in LLM intermediate activations, recoverable with a simple linear decoder without retraining or weight modification.
70
LLM Classification Is Feature Engineering
Argues LLM classifiers should feed a downstream logistic regression, yielding calibration, threshold control, and principled use of structured covariates.
32
Appears with
- MCPproduct · 2
- Brandsorganization · 1
- Mandiantorganization · 1
- AWSorganization · 1
- large language models (LLMs)product · 1
- SQL Serverproduct · 1
- OWASPorganization · 1
- IAM policiesproduct · 1
- covert-channel attackmalware · 1
- CyberStrikeAImalware · 1
- distributed LLM inferenceproduct · 1
- privacy-sensitive organizationsorganization · 1
- GTIGorganization · 1
- Googlevendor · 1
- xgboostproduct · 1
- Kasperskyvendor · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.