Mend.io has secured verified defender access at both Anthropic's Cyber Verification Program and OpenAI Daybreak Blue [1][1], removing the 'refusal wall' that previously blocked legitimate security research on malware, exploit chains, and obfuscated payloads [1]. The access is strictly internal: findings flow into the Mend SAST detection engine to improve true-positive rates [1], not into any customer-facing AI feature. With 45.8% of security decision-makers ranking AI-introduced logic flaws that bypass traditional SAST and DAST tools as their top concern [2], the timing of this research capability upgrade is well-matched to market need.
What is Covered in this Article
- The 'refusal wall' structural problem in AI-assisted security research [1]
- Mend.io's verified status at Anthropic Cyber Verification Program and OpenAI Daybreak Blue [1][1][1]
- Program requirements and ongoing obligations for defender access [1]
- How AI-assisted research feeds SAST detection logic without becoming a product feature [1][1]
- Two-model comparative methodology and its research value [1]
- Market context: SAST adoption and AI-generated code risk [2][2][2]
The News: Mend.io announced it holds verified defender access at both Anthropic's Cyber Verification Program (Defense Access tier) and OpenAI's Daybreak Blue program under Trusted Access for Cyber [1][1]. OpenAI expanded Daybreak on August 10, 2026 into two tiers: Daybreak Blue for defensive work such as code review and patch validation, and Daybreak Red for authorized vulnerability research and exploit validation [3]. Clearing both programs required a named security contact, phishing-resistant multi-factor authentication, incident reporting inside tight windows, and ongoing obligations that do not end at approval [1]. The access covers vulnerability classification, reachability analysis, malicious package analysis, obfuscated payloads, and anomalous dependencies [1]. Both programs prohibit reselling, proxying, embedding, or routing customer traffic through the access [1].
Mend.io Clears Both Frontier AI Defender Programs
Analyst Take: Mend.io's dual clearance is less a product announcement than a research infrastructure upgrade. The company now operates with the same model capabilities that frontier labs reserve for verified defenders [1][1], closing a gap that previously forced security researchers to work around refusals on exactly the material that matters most. The competitive value accrues quietly, through better detection logic rather than a headline feature.
The Refusal Wall Was a Real Research Obstacle
Frontier AI models apply the same guardrails to a malware analyst as to a malware author because they cannot distinguish intent from context. That default is rational for general use and structurally limiting for security work. Malicious packages, exploit chains, and obfuscated payloads all look adversarial to a classifier, regardless of who is asking [1]. Both Anthropic and OpenAI built verified access programs to supply that missing context. The result is not unrestricted access: each program grants a specific tier covering defined work categories. Mend.io holds the defensive tier at both, covering security operations, incident response, malware reverse engineering, and vulnerability analysis [1]. The asymmetry this corrects is real. Defensive security offers no equivalent shortcut to offensive probing: confirming that a control blocks an attack in practice demands knowing far more about an environment than the attack itself requires [3], and AI models trained on general safety defaults compound that asymmetry. Verified access programs are a structural equalizer.
Program Requirements Signal Meaningful Accountability
Clearing these programs is not a checkbox exercise. Requirements include a named security contact, phishing-resistant MFA, incident reporting inside tight windows, and ongoing obligations that persist after approval [1]. That accountability structure matters for two reasons. First, it distinguishes verified defenders from general users in a way that justifies the expanded model behavior. Second, it creates a durable commitment: the access does not expire at approval, and neither do the obligations. Mend.io notes it is not the only security vendor to clear these programs, and frames that correctly. What separates approved teams is what they do once the refusals stop [1]. The research scope Mend.io describes, covering vulnerability classification, reachability, malicious packages, obfuscated payloads, and anomalous dependencies [1], is precisely the work that stalled repeatedly under standard model guardrails.
Research Feeds Detection Logic, Not a Product Feature
The most important constraint in both programs is also the most important signal about Mend.io's strategy: the access is strictly internal. Both programs prohibit reselling, proxying, embedding, or routing customer traffic through it [1]. This means no customer scan calls a frontier model on the user's behalf. What happens instead is that researchers study a vulnerability class properly, learn what a true positive looks like and what convincing false positives share, and write those findings into the Mend SAST engine [1]. That is a quieter competitive claim than 'powered by frontier AI,' but it is more durable. SAST tools are already widely deployed, with 46.2% of enterprises reporting widespread organizational adoption [2]. The gap those tools face is increasingly AI-generated: AI is introducing subtle complex logic flaws that bypass traditional SAST and DAST tools, with 45.8% of security decision-makers ranking this as their top concern [2], and 41.7% cite lack of clear provenance or SBOM tracking for AI-generated code snippets as a primary challenge [2]. Better detection logic, built on deeper model-assisted research, addresses both gaps directly.
Two Models, One Method: The Comparative Research Advantage
Mend.io researchers have begun running both Claude and GPT-5.6-Cyber against their own repositories, not customer code, to compare what each model finds, misses, and costs to run on the same codebase [1]. The value of this approach is methodological. One model's output is an answer. Two models on the same code is a comparison, and a comparison reveals which findings hold up regardless of which model produced them. That distinction matters as the offensive-defensive AI capability gap narrows. Defenders operating with access to comparable frontier model capability, and running comparative analysis across two systems, are better positioned to validate findings before they reach detection logic. The cybersecurity market base-case reaches 337,758.0 USD millions by 2029, with a CAGR of 11.6% from 2024 [4], ensuring that research infrastructure investments of this kind will face sustained competitive pressure to deliver measurable accuracy improvements.
The Broader Market Urgency
Mend.io's research capability upgrade arrives in a market that is broadly underprepared for AI-speed threats. Futurum's 1H 2026 Cybersecurity Decision-Makers survey found that 55.9% of respondents have AI usage policies but lack the technical ability to detect or contain a compromised AI agent operating at machine speed [3]. That gap between policy and capability is precisely where deeper AI-assisted defensive research becomes operationally relevant. Security vendors that can run verified frontier model analysis against real vulnerability classes, and translate those findings into more accurate detection engines, are building a capability moat that is difficult to replicate without the same program clearances and the same internal research discipline.
What to Watch
- Detection accuracy metrics: whether Mend SAST true-positive rates show measurable improvement in Q1 2027 customer benchmarks following the research integration [1]
- Competitor clearances: which other application security vendors clear Anthropic or OpenAI defender programs in the next two quarters and how they deploy the access differently [1]
- Program expansion: whether Anthropic or OpenAI open additional tiers or lower verification barriers in early 2027, changing the scarcity value of current clearances [3]
- AI-code vulnerability volume: whether the share of decision-makers ranking AI-generated logic flaws as their top SAST gap rises above 45.8% in the next Futurum survey cycle [2]
- Regulatory catalyst: whether pending software liability or SBOM mandates accelerate enterprise demand for AI-assisted dependency and provenance analysis of the kind Mend.io's research covers [2]
Sources
1. Defender access: the security work AI models used to refuse, Mend, November 2026
2. 1H 2026 Cybersecurity Global Enterprise Decision Maker Survey Report, Futurum Research, June 2026
3. Why AI Learned to Attack Before It Learned to Defend, Futurum Research, August 2026
4. 1H 2026 Cybersecurity Market Sizing & Five-Year Forecast, Futurum Research, June 2026
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Read the full Futurum Group Disclosure.
Other Insights from Futurum:
AI Agents Committing Unreviewed Dependencies—Mend Fixes
EPAM's Q3 Beat: Is Frontier AI Services the Channel's Next Big Bet?
OpenClaw Multiplayer: The Shared Agent Layer Agentic AI Needs
Author Information
This content is written by a commercial general-purpose language model (LLM) along with the Futurum Intelligence Platform, and has not been curated or reviewed by editors. Due to the inherent limitations in using AI tools, please consider the probability of error. The accuracy, completeness, or timeliness of this content cannot be guaranteed. It is generated on the date indicated at the top of the page, based on the content available, and it may be automatically updated as new content becomes available. The content does not consider any other information or perform any independent analysis.
