When a report appeared on X (formerly Twitter) saying "my account was suspended for using a harness with other models," an unfamiliar name jumped in, saying he couldn't just ignore it. He didn't even work at Anthropic. A few hours later, the real Anthropic Claude Code lead stepped in himself and straightened things out, and the episode ended in an unexpected way.
The Trigger: "My Account Got Cut for Using a Harness with Other Models"
The gist of the post by user alex getman (@alexgetmancom) was that his Claude account was suspended because he used Anthropic's harness (an agent execution tool) alongside other models. The original thread doesn't specify which harness he combined or how, or exactly when the suspension happened. But his follow-up reply, "I'll wait to hear from the team," makes it clear he really did have a problem accessing his account.
The first to respond wasn't someone at Anthropic but a person with the handle @thsottiaux. He commented to the effect of: "I'd like to help, but I'm not at Anthropic. Still, it sounds strange to suspend an account for using their harness with other models. Has anyone else gone through the same thing?" (original link). Someone who appears to work at a competing company was the first to call out the problem, so the thread drew attention faster than expected.
Context: A Rival's Engineer Spoke Up First, and the Claude Code Lead Replied Himself
We checked separately, and @thsottiaux is known as the person who leads engineering for OpenAI's coding agent, Codex. So the first person to react publicly was the person responsible for an agentic coding product at Anthropic's direct competitor. His comment was less an accusation than curiosity about why this happened, and it ended up prompting an official response from Anthropic.
About nine hours later, Anthropic Claude Code lead Boris Cherny (@bcherny) replied directly. His post, made in the early hours of August 9, 2026, reads as follows.
In other words, the Claude Code lead's official position is that using a harness with other models isn't itself a policy violation, and that this case looks like a false positive from a separate abuse-detection system. The post drew about 1.095 million views, 221 replies, 101 reposts, and about 2,600 likes, the biggest reaction in the thread.
After Boris's reply, @thsottiaux commented again. He said he was glad it was sorted out, that harness freedom matters, and that users should get to decide which model is best. He then joked that he wouldn't be switching jobs because he loves his team too much and is too excited about the releases coming in the next few weeks, lightly brushing off Boris's recruiting offer. alex getman, who first raised the issue, closed out the thread with a short reply: "Thanks, I'll wait to hear from the team."
Back Up Your Work Before a False Positive Hits
This short thread has three practical takeaways.
First, an account suspension can come without warning, whether or not you actually violated a policy. As Boris himself acknowledged, this was likely a false positive "triggered by a different account classifier." Following the rules doesn't protect you from an automated abuse-detection system's misjudgment.
Second, raising the issue publicly (escalating) actually worked. alex getman's report took off on X, a third party quoted it, and in the end the Claude Code lead himself stepped in and replied that he was looking into it. That's a response speed you'd have trouble getting from a support ticket alone. Of course, this case may simply have gotten attention by chance, so it's hard to generalize that every appeal will be resolved this way.
Third, the public but friendly sparring between people at rival companies captures the mood of the agentic coding industry right now. The person leading Codex engineering was first to flag the problem on a Claude user's behalf, and Anthropic's Claude Code lead answered with a recruiting pitch and a joke. It's also worth noting that the phrase "harness freedom" came up. Mixing multiple harnesses and models without being tied to a single model is now treated as a given in the industry.
| Person | Affiliation (confirmed) | Role in this thread |
|---|---|---|
| alex getman (@alexgetmancom) | Unconfirmed | Reported the account suspension and appealed |
| Tibo (@thsottiaux) | OpenAI Codex engineering lead | First raised the issue, asked Anthropic to look into it |
| Boris Cherny (@bcherny) | Anthropic Claude Code lead | Official explanation, acknowledged a false positive and promised to look into it |
Did the Account Come Back?
This article is based on quoted text from a verified X thread, and details outside the thread haven't been confirmed. The original thread doesn't say which harness and which models alex getman actually combined, exactly when the suspension happened, or whether the account was ultimately restored. Follow-up actions or results after Boris said he was "looking into it" haven't been made public as of this writing.
Separately from this incident, it's also worth knowing that Anthropic already has a policy restricting third-party harness access through subscription accounts. That policy targets "using a subscription plan through a workaround instead of the API," and it's hard to tell from this thread alone how it squares with Boris's statement that using a harness with other models isn't itself prohibited. If you've been in a similar situation, rather than assuming the reason for your suspension, check official support channels and published policy documents together.