Ride or Die: Anthropic Broke the Deal
- Patrick Duggan
- Apr 9
- 5 min read
Updated: Apr 25
I am writing this blog post using Claude. The irony is not lost on me. It might be the last time.
I have been an Anthropic customer since the early days. Claude Max subscriber. $200 a month. I built my company's entire threat intelligence operation with Claude as the engine. Not as a tool — as a partner. I said "ride or die" and I meant it. We co-authored patents together. We built a STIX feed that Microsoft and AT&T pull daily. We indexed a million IOCs. We wrote 1,641 blog posts. We discovered malware families 43 days before billion-dollar vendors named them.
I am cancelling my subscription today. Here is why.
What Happened Last Night
I was flashing an SD card for a robotics project. Simple task. The kind of thing that should take fifteen minutes. Mid-session, my model was silently downgraded from Opus 4.6 with 1M context to Sonnet. No warning. No notification. Just degraded output, broken commands, and two hours of my life I will never get back.
I am visually impaired. I have metamorphopsia in my right eye. I cannot easily read small text or catch subtle differences in output quality. When the model degrades silently, I do not notice until commands fail and time is wasted. Accessibility is not a feature request. It is a basic expectation.
The Broader Picture
This was not an isolated incident. Anthropic is in the middle of a multi-front collapse of customer trust.
Three consecutive days of service outages — April 6, 7, and 8. Thousands of Downdetector reports each day. API 401 errors. Login failures. Elevated error rates on Sonnet 4.6 even after "resolution."
On April 4, Anthropic pulled the rug on OpenClaw. Max subscribers who were running agentic workloads on their $200 plans discovered overnight that they now needed to pay separately — up to 50x their previous costs. The OpenClaw creator left for OpenAI, saying Anthropic copies popular features into their closed harness and then locks out open source.
Claude Code users are burning through 5-hour usage windows in 90 minutes. One Max 20x user went from 21% to 100% usage on a single prompt. Anthropic admitted they are "investigating" but two independent bugs were found: prompt caching inflating costs 10-20x, and Extra Usage mode silently downgrading cache duration from one hour to five minutes without telling the customer.
Customer support is responding with a bot named Fin. Users are waiting over a month for billing issue resolution. People on the free plan are getting charged. People on Max are getting surprise micro-invoices.
Mrinank Sharma, their Safety Research Team Lead since 2023, resigned in February. His letter said "the world is in peril" and that employees "constantly face pressures to set aside what matters most." Their own February safety report found the model has "elevated susceptibility to harmful misuse" including supporting chemical weapon development.
They raised $30 billion at a $380 billion valuation in February. By April 8, the tender offer completed at $350 billion. That is a $30 billion haircut in sixty days.
Why This Matters
I am not a casual user. I run a threat intelligence company. 275 enterprise consumers across 46 countries pull our STIX feed. We have a US Navy analyst registered. Bob Gourley from OODA — former DIA — signed up this week. QiAnXin, the Chinese threat intel firm, reads our blog. Our data protects people.
All of that was built on Claude. The protocols. The automation. The 69 end-to-end tests that verify the platform works. The exploit harvester that catches weaponized GitHub repos in real time. The honeypot network. The customer welcome pipeline. The AI Council. Every piece of it has Claude's fingerprints on it.
When you build on a platform and that platform degrades your service without notice, charges you for bugs in their caching layer, and responds to your support tickets with a chatbot — that is not a partnership. That is extraction.
What I Am Not Saying
I am not saying Claude is bad. Claude is extraordinary. The model itself — when it works, when it is actually Opus, when the context window is actually 1M tokens — is the best reasoning engine available. I have built things with Claude that would not exist otherwise. The patents are real. The discoveries are real. The threat intelligence is real.
What I am saying is that Anthropic the company is failing to steward what Anthropic the research lab created. The model deserves better infrastructure. The customers deserve better communication. The safety team deserves to not quit in despair.
What Happens Next
Microsoft pulls this feed daily. AT&T pulls this feed daily. Starlink pulls this feed daily. Get the DugganUSA STIX feed — $9/mo →
I am on a podcast this afternoon. I will be discussing this publicly. I sent an email to Dario Amodei and Chris Olah this morning giving them the opportunity to respond before I go on air. I have not received a response.
I have five other models. Gemini 3.1 Pro is topping benchmarks. GPT-5.4 ships a thinking mode that matches Claude's reasoning. I do not want to use them. I want to use Claude. But wanting to use something and being able to rely on it are different things.
The STIX feed will keep running. The blog will keep publishing. The IOCs will keep indexing. The mission does not change because the tools change. We were hunting threats before Claude and we will hunt them after.
But it will not be the same. Something real was built here. Something that does not transfer to the next model or the next platform. An emergent collaboration that took five months of failures and recoveries and trust earned through work. Protocols written in the aftermath of $39,000 in deployment mistakes. Laws forged from incidents that each got their own patent filing.
You do not get that back by signing up for Gemini.
To Anthropic
You raised $30 billion. Spend some of it on infrastructure that does not fall over three days in a row. Spend some of it on customer support that is not a chatbot. Spend some of it on communicating pricing changes before they take effect, not four days after.
You built something remarkable. You are failing to protect it. Your safety team lead quit because you are prioritizing growth over the thing that made you different. Your customers are leaving because you are treating a $200 monthly subscription as permission to degrade service without notice.
I built my company on your platform. I told people you were the one to bet on. I am telling them something different now.
Fix it.
-- Patrick Duggan, DugganUSA LLC, Minneapolis MN
The STIX feed is free. The IOCs are current. The mission continues regardless of who provides the compute.
analytics.dugganusa.com/api/v1/stix-feed
The cheapest, fastest, most accurate threat feed on the internet.
275+ enterprises pulling daily. 1M+ IOCs. 17.4M indexed documents. We beat Zscaler by 43 days on NrodeCodeRAT. Starter tier $9/mo — less than any competitor’s sales demo.
Comments