Shrewd Research studies the limits of AI safety: which defenses can work, which cannot, and which assumptions fail under pressure. We publish evidence so policy, funding, and technical decisions do not rest on false claims.
The most valuable safety result can be negative: proof that a promising path cannot deliver the protection people expect from it.
Every measurement traces back to the metre — an independent standard no one gets to set for themselves. AI security has no equivalent: providers assert their own safety, and everyone else has to take it on trust. Shrewd is building that reference standard. Every claim about a model's containment, isolation, or integrity should trace to tamper-evident evidence an independent party can verify, replicate, and revoke. The business value is certainty you can act on: stop investing in defenses the evidence has already disproved, and stop trusting claims that cannot survive contact with a capable adversary.
Read the evidenceStudy the fundamental trade-offs and failure boundaries of proposed AI defenses. We ask what a control can guarantee, what assumptions it requires, and where those guarantees become impossible.
A rigorous result that rules out a bad direction, narrows the viable design space, or identifies the assumptions a defensible approach must satisfy.
Design experiments that pressure-test models, safeguards, evaluation harnesses, and AI infrastructure against adaptive attacks rather than static benchmarks.
Reproducible evidence showing where a system fails, how reliably it fails, and which conclusions the data does and does not support.
Translate an important safety or security claim into a falsifiable test, then evaluate it independently against realistic threats and deployment conditions.
A clear pass, fail, or conditional result with the evidence and assumptions needed for others to challenge or reproduce it.
Replicate influential findings, build harder evaluations, and test whether reported results survive new models, stronger adversaries, and changed environments.
An independent replication, open benchmark, or evaluation method that makes a disputed claim easier to verify.
Turn technical findings into precise implications for policy, standards, funding, evaluations, and infrastructure design without overstating what the evidence proves.
A decision memo, technical standard input, or evidence-backed recommendation that separates established facts from open questions.
The research becomes useful when it changes real systems. These five applications turn a safety claim into a test, preserve the evidence, explain failures, support independent evaluation, and permit assurance only while the evidence holds.
Attack the system and validate its security claims.
02_RECORDPreserve trustworthy evidence of what actually happened.
03_INVESTIGATEReconstruct failures automatically.
04_AUDITLet independent evaluators verify models securely.
05_ATTESTIssue an assurance result people can actually trust.
Every important fact about a model — provenance, training environment, hardware, checkpoints, evaluations, deployment configuration, permissions, safeguards and runtime events — becomes signed, tamper-evident evidence. All five products are different views over the same graph.
A multi-month containment failure mapped stage by stage against the assurance loop.
Open case fileMitigates tool squatting and rug pull attacks in the Model Context Protocol (MCP) by hardening tool definitions with OAuth-enhanced signatures and policy-based access control, preventing malicious or mutated tools from being silently substituted into agent workflows.
Introduces a hierarchy-based machine-learning approach to classifying file fragments — a core digital-forensics problem when file-system metadata is unavailable — improving type identification from raw binary content.
Investigates multiproperty coupling among magnetic field, piezoelectricity, and photoexcitation in ZnO nanowire arrays, demonstrating order-of-magnitude current enhancement through piezo-magnetotronic and piezo-photomagnetotronic effects.
Uses relocation entries in ELF binaries to fingerprint Linux kernel versions from memory images, enabling forensic tools to apply correct kernel profiles during memory analysis.
Describes a hands-on curriculum using virtual machine introspection to teach operating-systems security, letting students observe rootkits and live system state from outside the compromised guest.
Presents an automated approach to detecting and mitigating kernel rootkits in cloud environments using virtual machine introspection, without agents inside the guest.