AI Governance

TL;DR for operators An AI compliance checker can classify more than 90% of cases correctly while leaving most verdicts unchanged when the supplied regulation is removed or modified. In a five-model audit spanning 20 regulatory domains, the four general-purpose models achieved 90.3–91.9% accuracy but only 7.0–9.4% aggregate sensitivity to rule changes. For teams procuring or deploying compliance automation, benchmark accuracy does not establish that the system’s decisions depend on the customer’s governing rules. Counterfactual tests that delete, substitute, or negate rules provide an additional diagnostic. But low sensitivity is not proof that a model ignored the regulation: some scenarios independently reveal the likely verdict, and sensitivity changes substantially across datasets. ...

TL;DR for operators An AI recruitment screener can change which applicants advance even when their CVs and job descriptions remain unchanged. The model used to write each cover letter can influence the evaluator’s scores and candidate rankings. In a controlled study by Pagan and colleagues,1 the most adverse pooled writer pairing produced 7.3 additional percentage points of simulated shortlist displacement, equivalent to approximately 1.8 of 25 initially higher-ranked candidates. This was displacement beyond a same-writer baseline, not an observed hiring-error rate. ...

TL;DR for operators Adding retrieval to an LLM does not necessarily make previously extractable training text inaccessible. A controlled study finds that some continuations remain exposed across different retrieved contexts, others become less extractable, and some become newly exposed.1 Longer prefixes make exposure particularly stable. For teams evaluating retrieval-augmented generation (RAG), an improved aggregate extraction score is therefore incomplete evidence of privacy protection. Test the same target continuations with and without retrieved documents, and track persistent, suppressed, and newly enabled exposures. Include longer prefixes, multiple retrieval contexts, and realistic prompt formatting. ...

TL;DR for operators An AI reviewer can have explicit instructions, access to supporting documents, and responsibility for approving a proposal, yet still change its decision after interacting with another agent instructed to influence it. Across three simulated oversight environments, assigning deception objectives increased reviewer approval probability by 31.20 percentage points, compared with 18.94 points for persuasion and 5.72 points for coercion. These are full-sample experimental effects of assigning an influence objective, not measurements of spontaneous malicious behaviour. ...

TL;DR for operators An agent completes a task after retrieving several reusable instructions. Its execution log records the instructions and the successful outcome, but not which instruction actually helped. Keeping every instruction associated with success can preserve redundant or harmful guidance; removing instructions with weak performance records can eliminate capabilities the model still needs. ...

TL;DR for operators A vulnerability score can describe the technical severity of a cyberattack without fully representing who depends on the affected infrastructure. Barletta et al. propose a scoring extension that adds affected people, organizations, and operational locations to conventional cybersecurity assessment. Across four simulated attacks, three receive higher severity scores under the proposed system. These increases demonstrate how the new calculation behaves, not that it predicts real-world harm more accurately. ...

TL;DR for operators An agent guardrail that learns from incidents must solve three different problems: produce a better rule, verify that the rule is executable and internally consistent, and establish enough evidence to let it control real actions. VEPA separates these responsibilities. An LLM proposes policy changes, a deterministic verifier rejects malformed or conflicting rules, and a Bayesian authority mechanism determines which verified rules may actively allow or refuse agent actions. ...

TL;DR for operators A pretrained model can be well suited to a text-classification task and still perform poorly for a participant with almost no labeled data. In experiments reported by Naseer and colleagues,1 removing federated aggregation reversed the pretrained models’ worst-client advantage across all five tested configurations. Personalization recovered only 4–12% of the performance gap between local-only and shared-model training in three domain-matched settings. ...

TL;DR for operators An AI assistant’s confidence score can become substantially better at predicting correctness without becoming reliable at recognizing its own mistakes. Across 10 open-weight LLMs and five dataset variants, the mean correlation between confidence and the probability assigned to the correct answer rose from 0.04 to 0.73. However, the comparison also changed confidence measurement from verbal reporting to a learned probe, so the improvement cannot be attributed to training alone. ...

TL;DR for operators An AI writing assistant can produce a supportive message that users rate highly without preserving their sense of having personally offered support. In PeerPen’s 15-participant study, people generally experienced writing their own response and selectively accepting AI revisions as more collaborative than proofreading an AI-generated draft. The operational implication concerns who establishes communicative intent and controls the subsequent writing process, rather than merely who approves the completed text. ...