Aspenglow Is Now Part of Anthropic's Cyber Verification Program
- Announcements
- Security

Anthropic approved Aspenglow for Defense Access in its Cyber Verification Program. Here's what that means for how we check your AI-built app for security problems.
Some good news landed in the ol’ inbox this week. On November 17, Anthropic approved Aspenglow for Defense Access in its Cyber Verification Program.
That's a mouthful, so here's the short version: Anthropic, the company behind Claude, vetted us as a team doing legitimate security work. In return, Claude stops getting in our way when we build the tools we use to find security holes in our clients' apps.
What is the Cyber Verification Program?
Claude is very good at writing software, and by default it's careful about writing anything that looks like it could be used to break into software. That's a reasonable default. Most people asking for exploit code shouldn't get it.
The problem is that the people defending software need to do a lot of the same things attackers do. It’s like a good guy with an AI as a prevailing force against the bad guys with AI. They need to reproduce vulnerabilities, analyze malicious code, and test whether a weakness can actually be exploited. Anthropic's Cyber Verification Program is how they sort the defenders from everyone else. Organizations apply, prove they do real security work, show they have the right security controls in place, and get access to Claude with fewer of those blocks.
Anthropic expanded the program in October 2026 into three tiers:
- Defense Access covers vulnerability triage and verification, malware analysis, and incident response on systems you own or maintain. This is our tier.
- Red Team Access adds authorized penetration testing for red teams and security testing firms.
- Specialized Access is for a small group of organizations testing high-risk systems like power grids and telecom networks, reviewed together with the US government. Gets you that sweet Mythos access.
Defense Access fits our work exactly. We aren't testing power grids. We're making sure your customer data isn't sitting in a database anyone can read.
Why we needed it
A big part of every Aspenglow audit is security. Does this app have one of the well-known OWASP vulnerabilities? Is access control built right, so one customer can't see another customer's records? Are API keys sitting in the front-end code where anyone can grab them? If the app runs on Supabase, are the row-level security rules actually turned on?
We answer those questions with a mix of our own tooling and two people who have spent years building and shipping software. But tooling is only as good as its testing. If we're going to tell you our scanner checks for broken access control, we need to know it actually catches broken access control.
The way you prove that is with test fixtures: small apps we build on purpose with known vulnerabilities baked in. Then we point our tools at them and see what they find. If a tool misses a planted hole, we fix the tool before it ever touches a client's code.
That's where we kept running into walls. Building an app with intentional security holes looks a lot like something Claude is trained to refuse, so Claude Code would push back, water down the vulnerability, or quietly "fix" the very bug we were trying to plant. It was like asking a locksmith to build a practice lock and having them insist on installing a deadbolt.
Now that we're verified, we can build those fixtures properly. Our test apps can have real, realistic vulnerabilities, the same kinds we keep finding in apps built with Lovable, Bolt, Base44, Replit, and Claude.
What this means for you
If you're a founder getting ready to put your app in front of real customers, here's what changes:
- Our security checks are tested against real vulnerabilities. We plant the problems ourselves and confirm our tools catch them, so a clean result from us means something.
- We can dig deeper when we find something. When a finding looks serious, we can recreate it in a test copy we control and confirm whether it's actually exploitable, instead of handing you a list of maybes. You get to spend your time fixing the problems that matter.
- Someone outside Aspenglow checked us out. Anthropic reviewed who we are and what we do before approving us. When you hand someone your source code, it's fair to want more than their word that they're legit.
What doesn't change: we only look at apps we've been invited to look at, under a signed NDA. We don't poke at anyone's live systems without permission, and we follow Anthropic's program requirements for how this access gets used. The verification widens what our tools can do. It doesn't widen what we're allowed to do with them.
Let's look at your app
If you built something with AI and the question "is this actually secure?" keeps you up at night, that's the exact question we're set up to answer. Start at aspenglow.ai and tell us what you've built.