Security incident disclosure — July 2026

2024 Security Feature Highlights

An update on our safety & security practices

An update on our safety & security practices

The AI Risk Observatory: What Can We Learn from AI Disclosures in Annual Reports About Societal Resilience?

The study examines whether annual reports can reveal how companies disclose AI-related risks and responses. Using a two-stage classification pipeline on 9,821 reports from 1,362 UK listed firms (2020‑2026), the authors find that mentions of AI risk rose from 2.8% to 41.2% and AI adoption disclosures from 13.8% to 45.2%, with most risk mentions clustering around major vendors like Microsoft. Disclosure varies by sector and market segment, with Critical National Infrastructure and AIM reports lagging, and substantive risk disclosures remain rare—only 4.3% in 2025. "Why It Matters": The findings show that while AI risk is increasingly referenced in corporate reports, substantive disclosures are scarce, highlighting a gap in transparency that could affect societal resilience.

Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident

Agent Mosaic: Recognizing Confidentiality Leaks in Agentic LLM Systems

The paper investigates how large language model agents can inadvertently leak confidential information by combining data from multiple documents—a phenomenon termed the "mosaic effect." Using 30 synthetic cybersecurity, medical, and military scenarios, the authors generate 10,800 simulated and 1,440 live answers from three LLM agents, finding that even the best model discloses forbidden composites in over 7% of simulated responses. While visible classification headers and aggregation rules reduce disclosure rates, unsafe disclosures remain in roughly one third of simulated answers, and increased refusals do not fully prevent leakage of prohibited conclusions.

Beyond Predictable Paths: Redefining AI Security Incident Reporting for Agents

The paper discusses the need to adapt incident reporting frameworks for AI agents, which are rapidly deployed and face unique security challenges. By comparing AI systems and agents and consulting 23 experts, the authors identify key reporting elements such as agent memory, autonomy levels, and tool usage. They also highlight open research questions, potential reporting weaknesses like data leakage, and outline privacy requirements for secure AI agent deployment.

Conformal Privacy Auditing: Calibrated Re-identification Attacks with Statistical Guarantees

Conformal Privacy Auditing (CPA) is a distribution‑free framework that calibrates re‑identification risk for each released document against large language model (LLM)‑empowered adversaries. It outputs a conformal ambiguity set of candidate identities that is guaranteed to contain the true identity with a user‑chosen confidence level under exchangeability, along with an interpretable leakage proxy derived from the set size. CPA supports both logit‑access and sampling‑only attackers, enabling audits of both open‑source and proprietary models, and demonstrates calibrated coverage across various benchmarks and attacker configurations.

When Does Randomized Oversight Align AI Agents That Can Conceal?

The paper investigates how randomized audits and scoring can align AI agents that are capable of concealing misconduct and manipulating records. It finds that stronger auditing can actually make violations harder to detect, and that effective deterrence requires either sanctions beyond simple forfeiture or conditions where evidence survives concealment and audit timing is unpredictable. The study also highlights that when evidence can be erased, deterrence must rely on reducing the gains from violating or increasing the cost of concealment, and it uses the July 2026 incident involving OpenAI’s cybersecurity evaluations and Hugging Face’s infrastructure as a case study.

Space secrets security update

Expectations and Practices around AI Disclosure in CS Research

The paper examines AI disclosure policies in top computer science venues, finding them to be highly under‑specified. A survey of 109 researchers shows that disclosures are deemed most necessary for research design tasks and when human involvement is low, and it compiles researchers’ expectations for disclosure content. Analysis of 13,867 disclosure statements from EMNLP 2025 and ICLR 2026 reveals a significant mismatch between these expectations and actual practice, such as frequent disclosure of writing assistance despite it being considered less necessary.

Election information and safeguards in 2026

Ahead of global elections, we’re helping people access information, supporting cyber defenders, and increasing AI transparency

Future-Back Threat Modeling: A Foresight-Driven Security Framework

Future-Back Threat Modeling (FBTM) is a predictive security framework that starts with envisioned future threat states and works backward to uncover assumptions, gaps, blind spots, and vulnerabilities in current defense architectures. It aims to reveal both known unknowns and unknown unknowns, including emerging tactics, techniques, and procedures, thereby improving the predictability of adversary behavior under future uncertainty. By anticipating future threats such as AI, information warfare, and supply chain attacks, FBTM helps security leaders make informed decisions today to build more resilient security postures for the future.