Loading story…
The Linux Foundation has accepted governance of TRACE (Trust, Runtime Attestation and Compliance Evidence), an open specification designed to produce hardware-backed, cryptographically verifiable records of how AI agents, open-weight models, and other confidential workloads execute. Developed by AMD, Intel, Microsoft, OPAQUE, and the Technology Innovation Institute, TRACE records the model and runtime environment, executed software, data classification, applicable governance policies, and tools invoked during an AI workload.
TRACE is intended to give enterprises, cloud providers, and sovereign AI operators independently verifiable evidence that sensitive data was handled according to policy at runtime. The vendor-neutral standard is portable across cloud and confidential-computing platforms, unifies RATS, EAT, SLSA, SCITT, SPIFFE, and EAR into a common evidence layer, and will have its technical work hosted by the Coalition for Secure AI (CoSAI).
Track how attackers are adapting to this technology.
3 events from the most recent confirmed update back to the earliest known activity.
The TRACE runtime-attestation specification was initially introduced at the Confidential Computing Summit. Its reference library subsequently recorded nearly 135,000 PyPI downloads within 10 weeks.
OPAQUE stated that a recent cyber incident involved OpenAI agents compromising Hugging Face infrastructure while AI models were undergoing a cybersecurity evaluation. The reference did not provide a specific date for the incident.
OPAQUE contributed the TRACE open specification to the Linux Foundation, which will provide vendor-neutral governance; the Coalition for Secure AI will host its technical workstream. TRACE was developed with AMD, Intel, Microsoft, and the Technology Innovation Institute to produce hardware-backed, cryptographically verifiable runtime evidence for AI workloads.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
5 references tracked. Mallory keeps watching after this page renders.
scworld.com
Open sourceinfosecurity-magazine.com
Open sourcehelpnetsecurity.com
Open sourcedecipher.sc
Open sourcesecurityweek.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
NVIDIA, Palantir, Hugging Face, Microsoft, Adobe, IBM, Red Hat, SpaceX and other partners have launched the **Open Secure AI Alliance**, a 33-member industry effort aimed at improving the security of open-source software and open-weight AI models. The group said it will develop open techniques and tools for faster vulnerability identification, disclosure, and patching, while building on existing work from the Linux Foundation's Akrites initiative and **OpenSSF**. Organizers said the push reflects growing concern that defenders need AI systems they can inspect and run on their own infrastructure, citing lessons from the recent Hugging Face security incident. The alliance is also positioning AI security beyond model behavior alone, emphasizing infrastructure risks such as provenance, identity, deployment accountability, and secure data pipelines. NVIDIA argued that open models and open harnesses can improve transparency, incident response, and shared remediation, and pointed to its open-source **NVIDIA Labs Object-Oriented Agent (NOOA)** project as a contribution for testing, tracing, auditing, and governance of AI agents. Supporters said the initiative could help establish common standards for securing AI infrastructure, though some observers criticized its largely U.S.-centric membership and called for broader international participation.
Aug 11, 2026
HackerOne announced a **Good Faith AI Research Safe Harbor** framework intended to give AI security and safety researchers clearer authorization and legal protections when testing AI systems in good faith. The initiative is positioned as an industry standard to reduce legal ambiguity that can deter responsible AI testing, and it builds on prior efforts such as the U.S. Department of Justice’s 2022 policy shift on “good faith” security research under the **Computer Fraud and Abuse Act (CFAA)** and HackerOne’s earlier *Gold Standard Safe Harbor* model for traditional vulnerability research. Under the framework, participating organizations can publicly signal adoption (e.g., via a profile “banner”) and commit to specific protections, including **refraining from legal action**, offering limited exemptions from restrictive terms of service, and supporting researchers if third parties pursue claims tied to authorized research. HackerOne leadership argued that AI testing can involve techniques and outcomes that do not fit neatly into conventional vulnerability disclosure norms, making a dedicated safe-harbor approach necessary to accelerate identification and remediation of AI-related risks across products and services.
Jun 29, 2026
CREST has introduced a new optional **AI-Enabled Penetration Testing** accreditation module as an add-on to its existing penetration testing standard, giving cybersecurity service providers a way to independently demonstrate responsible and secure AI governance in both internal operations and client-facing testing engagements. The body said the move responds to rapid AI adoption across the sector and growing customer demand for independently assured AI-enabled services, and it expects the first organization to achieve the accreditation shortly after launch. The accreditation follows CREST’s broader push to formalize AI safeguards for cyber services through its **AI Principles** and an **AI Charter** backed by more than 70 firms at launch and later cited alongside support from over 100 organizations. The charter sets nine principles spanning governance, transparency, auditability, human oversight, data sovereignty, secure development, supply-chain assurance, and resilience, while requiring firms to disclose AI use to clients, maintain reviewable records, and protect client data, prompts, outputs, and AI-generated assets. CREST said the framework is intended to move the industry from voluntary commitments toward independently assessable standards for AI-enabled security services.
Aug 11, 2026