Effective September 29, 2026
This disclaimer applies to all research, articles, insights, reports, figures, code samples, benchmarks, and any AI-assisted output published by Devence Lab, Inc. ("Devence Lab", "we", "our", or "us") on devencelab.com or elsewhere (the "Materials"). It forms part of our Terms of Use.
1. Informational purposes only
The Materials are published for informational and research purposes only.
They are not, and must not be relied upon as:
- legal advice — on the EU AI Act, the GDPR, NIS2, DORA, sectoral regulation, contract, or liability;
- regulatory or compliance advice — on whether a system, control, or process satisfies any obligation applicable to you;
- security advice — on the architecture, configuration, hardening, monitoring, or assurance of your systems;
- audit, certification, or assurance of any product, vendor, model, or control;
- financial, investment, or procurement advice — on any company, product, security, or purchasing decision;
- medical, safety-critical, or engineering advice; or
- professional advice of any other kind.
No lawyer–client, consultant–client, auditor–client, or other professional relationship is created by your reading the Materials, by your contacting us, or by our responding to you, and we owe you no duty of care in respect of any decision you take on the basis of them. You should obtain advice from a qualified professional, retained by you, before acting on anything you read here.
2. Your systems are not our subject
Our research describes systems, failure modes, attack techniques, and mitigations in general terms, on the basis of published sources, reproducible experiments, and our own engineering work.
It is not an assessment of your systems, your threat model, your regulatory position, or your risk. A finding that holds for the system we tested, under the conditions we tested it, may not hold for yours. A control that was adequate in a published evaluation may be inadequate in your deployment. Validate independently before acting.
3. Sources, methods, and figures
We cite our sources and, where we can, link to primary material. Our Editorial Policy describes how we source, review, correct, and update published work.
Where we report a quantitative result — a benchmark score, a detection rate, a latency figure, a cost — it reflects the methodology, configuration, model version, dataset, and date stated in the piece. Results of this kind are highly sensitive to conditions that change quickly: a model version, a system prompt, a decoding parameter, a dataset revision, or a provider-side change can move a figure substantially without any announcement. Treat every such figure as a measurement taken at a moment, not as a property of the system.
Where we reproduce a third party's result, we attribute it and do not warrant it. Where we describe a vendor's product, we describe it as we found it on the date stated, and we do not speak for the vendor.
4. Currency
The Materials describe the state of affairs at the date of publication. AI systems, the products built on them, the attacks against them, and the law governing them all change rapidly. A piece that was accurate when published may be substantially out of date within months.
We do not undertake to update published work to reflect later developments, except where our Editorial Policy requires a correction. Always check the publication date, and seek current information before relying on anything time-sensitive.
5. AI-assisted content and output
Some of our Materials, and some features we may make available, are produced or assisted by artificial intelligence systems.
AI-assisted content and AI output may contain errors. These systems can produce statements that are plausible, fluent, well-structured, and wrong. They can fabricate citations, invent sources, misattribute quotations, and misstate numbers with complete confidence. They can reproduce biases present in their training data.
Accordingly:
- Every AI-assisted output must be independently reviewed and verified by a competent person before any operational, commercial, legal, or safety-relevant use.
- Verify every citation against its primary source before relying on it or repeating it.
- Do not use AI output as the sole basis for a consequential decision.
- Do not deploy AI output into a production, safety-critical, or regulated environment without review, testing, and appropriate controls.
- Where we make an assistant available on the Site, its answers are generated automatically, are not reviewed by a person before you see them, and do not represent our considered position. Do not enter confidential or sensitive information into it — see Section 4.2 of our Terms of Use.
How we source, review, and correct published research is set out in our Editorial Policy.
6. Security technique and dual-use material
Our research describes attack techniques, exploitation paths, and the limits of defensive controls, because a defence cannot be evaluated without describing what it defends against. This material is published so that defenders can understand and mitigate the risk.
Using it against systems you are not authorised to test is prohibited by our Acceptable Use Policy and is a criminal offence in most jurisdictions. You are responsible for the lawfulness of what you do with what you read.
Proof-of-concept code, where published, is illustrative and is provided without warranty of any kind. It is not production code, has not been hardened, and may be unsafe to run outside an isolated environment.
7. Forward-looking statements
The Materials may contain statements about expected developments, research directions, product plans, or industry trends. These reflect our view at the date of publication, are subject to risks and uncertainties, and are not commitments. Actual outcomes may differ materially. Nothing in the Materials is an offer, a solicitation, or a commitment to deliver any feature, product, or timeline.
8. Third-party names and links
References to third-party products, services, companies, models, or trade marks are for identification and commentary only. Those marks belong to their respective owners. Such a reference does not imply endorsement, affiliation, sponsorship, or partnership in either direction.
External links are provided for reference. We do not control and are not responsible for third-party content, security, or privacy practices. See our Editorial Policy.
9. No warranty and limitation of liability
The Materials are provided "as is", without warranty of any kind, express or implied, including any warranty of accuracy, completeness, currency, merchantability, fitness for a particular purpose, or non-infringement.
To the fullest extent permitted by law, we accept no liability for any loss or damage arising from reliance on the Materials, including any decision made or not made, any action taken or not taken, or any system deployed or not deployed on the basis of them. Sections 7 and 8 of our Terms of Use apply in full to the Materials.
10. Contact
Corrections, factual disputes, and questions about our research may be sent to contact@devencelab.com. We take correction seriously; our Editorial Policy explains how we handle it.