P056 Secure boot and firmware integrity
Can an untrusting verifier confirm an accelerator runs only signed, non-rolled-back, auditable firmware, and has secure boot been bypassed in published work?
Secure boot, anti-rollback and auditability of accelerator firmware.
Status
Analyzed Design options only: jointly created firmware reference measurements, and listed ways to attest that verification firmware stays present (secure boot made impossible to disable, a secure counter, randomly timed low-latency attestation); Shavit 2023 proposes signed-firmware enforcement, anti-rollback and early remote attestation of updates. Nothing built or tested. Bypass clause at a lower stage: no held document reports a secure-boot bypass of a data-centre GPU; flexHEG Part 2 describes voltage-glitch attacks on secure boot in general from a source not held; Skorobogatov & Woods 2012 broke an FPGA's protections, not an accelerator's. H100 secure boot and anti-rollback features are reported second-hand (Petrie 2024; Scher & Thiergart) from NVIDIA documents not held.
To obtain reliable reference measurements when the hardware designer’s measurements are untrusted, without gaining access to private information such as the Prover’s code, one option may be for the Verifier and Prover to jointly create reference measurements by emulating a compliant bootup on a confidentiality-preserving technology (e.g., Confidential Computing).
Posed by
- Baker, Kulp, Marks, Brundage & Heim (2025), Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment (RAND WR-A4077-1):
In secure boot, firmware attests to itself, so malicious firmware might hide its own malicious properties
(App. A.1 fn 86, p. 38, L2320)
Also raised in: Harack et al. (2025), Verification for International AI Governance (Oxford AIGI); Scher & Thiergart (2024), Mechanisms to Verify International Agreements About AI Development (MIRI); Petrie & Aarne (2025), Technical Options for Flexible Hardware-Enabled Guarantees (flexHEG Part 2); Aarne, Fist & Withers (2024), Secure, Governable Chips (CNAS); Shavit (2023), What Does It Take to Catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring; Hausenloy & Li (2026), The Compute Verification Post; Proofworks, Living Verification Documents.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P056.1 | Has secure boot on a current data-centre GPU been bypassed (e.g. voltage glitching, boot-ROM vulnerability) in published work, at what access level and equipment cost? | partly answered |
| P056.2 | Can an operator downgrade signed GPU firmware past the anti-rollback counter, and does any accelerator give an attestable, timestamped firmware-version history? | open |
| P056.3 | Of published accelerator security vulnerabilities, what fraction were fixable by firmware update? | open |
| P056.4 | Can an untrusting Verifier audit accelerator firmware (source + reproducible build matching the signed binary), so chipmaker–Prover collusion on backdoored firmware is excluded? | partly answered |
| P056.5 | Has scanning of boot-ROM / first-stage bootloader storage detected a planted firmware trojan that remote attestation missed? | open |
| P056.6 | Can reference measurements for firmware attestation be generated jointly (e.g. emulated compliant boot in a confidential environment) without trusting the designer? | open |
| P056.7 | Has a mechanism that keeps verification firmware continuously present (non-disableable secure boot, increment-only counter, random low-latency attestation) been implemented and red-teamed, closing the gap after the attestation moment? | open |
| P056.8 | Can a measured-boot chain on H100 CC be extended to the framework/model layer with a stated overhead? | open |
Sub-questions not yet stated precisely enough to check
| ID | Sub-question |
|---|---|
| P056.9 | Does formal verification of security firmware/hardware hold up given flawed specs, compilers or hardware models (cf. exploitable bugs in formally verified vendor code)? |
| P056.10 | Does AI-assisted vulnerability finding shift the offense–defense balance for chip security mechanisms? |
Possibly relevant work found by automated search
Found by AI agents searching for this problem's sub-questions, mostly by keyword and citation in the OpenAlex scholarly database and on the web; a second AI pass reviewed each one, and we spot-checked some by hand. They seemed reasonably related, as far as one can tell from outside each paper's field. Some come from neighbouring fields, such as arms control and financial audit, and bear on the problem by analogy. Not reviewed against the status above; a paper listed here does not change that status. 928 of 1052 quotes checked word for word against the paper's text.
For P056.1: Has secure boot on a current data-centre GPU been bypassed (e.g. voltage glitching, boot-ROM vulnerability) in published work, at what access level and equipment cost?
For P056.2: Can an operator downgrade signed GPU firmware past the anti-rollback counter, and does any accelerator give an attestable, timestamped firmware-version history?
For P056.3: Of published accelerator security vulnerabilities, what fraction were fixable by firmware update?
For P056.4: Can an untrusting Verifier audit accelerator firmware (source + reproducible build matching the signed binary), so chipmaker–Prover collusion on backdoored firmware is excluded?
For P056.6: Can reference measurements for firmware attestation be generated jointly (e.g. emulated compliant boot in a confidential environment) without trusting the designer?
For P056.7: Has a mechanism that keeps verification firmware continuously present (non-disableable secure boot, increment-only counter, random low-latency attestation) been implemented and red-teamed, closing the gap after the attestation moment?
* Full text not reviewed; more likely to contain errors.
Data
- problems.csv — one row per problem
- problem_status.csv — status, scope and evidence per problem
- problem_sources.csv — the posed-by quote and other documents per problem
- leaves.csv — all sub-questions, including those without a checkable form
- leaf_sources.csv — every source row behind each sub-question