Data Ingenious now has validated access to Claude Mythos 5.2, Anthropic's most capable cybersecurity model. We pair it with 25+ years of government-grade engineering to find, validate and fix security flaws in your software, including the old code nobody wants to touch.
Data Ingenious Global Limited, a Jaipur-based software company with 25+ years of government and enterprise delivery, offers this service. Our AI cybersecurity service finds security flaws in your software, confirms they are real, and helps fix them. After agreeing a written scope with you, we point Claude Mythos 5.2, a frontier AI model built for vulnerability discovery, at the code and systems you have authorised. It reads at machine speed and produces candidate findings. It does not make the decisions: a Data security engineer reproduces every finding, discards false alarms, rates the severity, and our engineering team can then patch and retest. The model runs on Anthropic's cloud under Anthropic's terms, so we agree up front what may be sent to it, and code that cannot leave your network is reviewed another way, using on-premises models and manual expert review.
Security review of your code and systems, with every finding verified, ranked by severity and carried through to a fix.
A written scope first. Then AI-assisted discovery, human validation, a prioritised report with fixes, and a retest.
Claude Mythos 5.2 reads in-scope code and suggests candidates. People confirm them. Code that cannot leave your network is reviewed on-premises.
Related briefing: what CERT-In and SEBI said about AI tools like Claude Mythos, and a 30-day checklist for Indian organisations.
Anthropic's published research on Claude Mythos Preview, the model line that Mythos 5.2 continues, reports results that changed what defenders should expect from AI. These figures are Anthropic's own, from its April 2026 research post.
Also reported by Anthropic: partners in its cyber programs found at least 129,000 verified vulnerabilities between April and July 2026. Sources: Claude Mythos Preview research, Project Glasswing, Cyber Verification Program. Figures are vendor-reported and reflect Anthropic's test conditions, not a guarantee of results on your systems.
Models with this capability will reach attackers as well as defenders. Organizations that scan and fix their own code first reduce the gap. Here is what that changes in practice.
Legacy systems, vendor code and forgotten internal tools are rarely audited because manual review is slow. A model that reads code at machine speed makes it affordable to check the code nobody has looked at in years.
Anthropic reports a single OpenBSD finding for under $50 per run, and about a thousand runs across that project for under $20,000. Broad coverage that once needed weeks of specialist time becomes a scoped, budgeted exercise.
The backlog problem is as much about noise as about discovery. Validated findings with severity and a reproduction path let your team spend time on what is real.
A report is not a fix. Because Data also runs modernization and product engineering teams, we can carry findings through patching, regression testing and release.
Regulated sectors need documented testing. Scoped, repeatable assessments with human sign-off give auditors and boards something concrete, including for AI systems you are rolling out.
AI agents add new risk: prompt injection, over-broad permissions, data leaving the building. We assess those as part of the same engagement, drawing on our own AI product work.
Each service is scoped in writing, run by Data security engineers, and uses Mythos 5.2 where it adds speed and depth.
Mythos-assisted review of your applications and services for exploitable flaws, with engineer-validated findings and fixes.
Deep review of old codebases, open-source dependencies and third-party components, linked to our modernization practice.
Turn scanner and bug-bounty backlogs into a ranked, reproduced, de-duplicated list your engineers can act on.
Augment penetration tests and red-team exercises on systems you own or are authorized to test, inside an agreed rules-of-engagement.
Threat modeling and testing for LLM apps and agents: prompt injection, tool permissions, data leakage, audit trails.
Patch, refactor or rebuild vulnerable components with regression tests, then retest. Find and fix under one team.
We agree targets, rules of engagement and written authorization. Nothing outside scope is touched.
Together we decide what may go to a cloud model and what must stay in your environment, and set the approach to match.
Mythos 5.2 reads and probes the in-scope code and systems at scale, generating candidate findings.
A Data security engineer reproduces each candidate, discards false positives, and rates severity and business impact.
You receive a prioritized report with reproduction steps and fixes. We can implement the patches with you.
We retest fixes and hand over evidence your auditors and leadership can use.
Written scope on every engagement. Testing only on systems you own or are authorized to test. Human validation of every finding. Coordinated, responsible disclosure for any third-party flaw we discover along the way. No offensive work against third parties, ever.
Claude runs on Anthropic's infrastructure and Anthropic applies its own safeguards, monitoring and data-retention terms to cyber access. If your code cannot leave your network, we say so up front and use on-premises models such as ZenithAI and manual review for those parts.
Tell us what you want assessed. A security engineer, not a sales queue, replies to scope a first engagement.
Prefer email? business@data.in
Data's AI cybersecurity service is the security practice of Data Ingenious Global Limited. It uses Anthropic's Claude Mythos 5.2 to find and validate vulnerabilities in source code and systems, with every finding reproduced by a human security engineer and fixes delivered by Data's engineering teams. Engagements cover only systems the client owns or is authorized to test.
Conventional scanners match known patterns. Anthropic reports that Mythos Preview reasons about code the way a security researcher does, finding flaws that had survived decades of review and automated testing, and building working exploits to confirm they are real. Data adds human validation so you receive reproduced, prioritized findings rather than raw alerts.
Government bodies, banks and fintechs, telecom operators, software product companies and enterprises with large or legacy codebases, in India and internationally. Engagements start with a written scope and authorization.
Claude Mythos is a class of Anthropic frontier models with the strongest cyber capabilities Anthropic has released, offered under restricted access to vetted security teams. Its April 2026 preview showed it finding and exploiting serious vulnerabilities in widely used software, including bugs that had survived decades of human review and automated testing.
Yes. Data Ingenious has validated access to Claude Mythos 5.2 and uses it, under Anthropic's usage terms, to support authorized security work for clients. Data is also a Registered member of Anthropic's Claude Partner Network. Data is an independent company and is not Anthropic.
Mythos-assisted source code security review, legacy and open-source dependency audits, vulnerability triage and validation, support for authorized penetration tests and red-team exercises, security of AI deployments (prompt injection, agent permissions, data exposure), and fix delivery through our modernization engineering teams.
Claude models run on Anthropic's cloud, so code or data sent to them is processed under Anthropic's terms, and organizations in Anthropic's verified programs are subject to data retention requirements. We decide this with you at scoping. Where code cannot leave your environment, we scope the work differently, using on-premises models such as ZenithAI and manual expert review for those components.
No. Every engagement starts with a written scope and authorization. We only test systems you own or are formally authorized to test. We do not offer offensive services against third parties.
Yes. Model output is a lead, not a verdict. A Data security engineer reproduces and validates each finding before it reaches your report, and rates its severity and business impact.
Use the form on this page or write to business@data.in. Tell us what you want assessed and any compliance or data-handling constraints. A security engineer replies to scope a first engagement.