I would assume they would keep that version of Claude for themselves internally, and not the public version of Claude. Why give their competitors any edge at all?
I think the answer is that they don’t give their competitors access. The Terms of Service explicitly prohibit using Claude to build a competing model or service:
D.4. Use Restrictions. Customer may not and must not attempt to (a) access the Services to build a competing product or service, including to train competing AI models[...]
And they’ve cut off access for both OpenAI and xAI because of this.
There’s limits to how enforceable this is if users are motivated enough, but apparently they feel like this is sufficient.
The alternate strategy of making multiple frontier models would be very expensive.
I doubt lack of official access is a real issue for anyone motivated. Honestly, what keeps people from xAI or OpenAI from still using Claude Code? Getting personal access? Is Antrophic actively monitoring for anything OpenAI related and blocking those accounts?
The whole point of using an AI coding agent is to reduce the amount of effort involved in coding. If you have to jump through hoops to ensure that your usage never shows up on a company IP, is it actually worth the effort (especially since OpenAI has their own models)? Plus a lot of people won’t pay out of pocket just to help the company (even if it would obviously be worth it), and the company would have to consider legal risks to intentionally violating the ToS.
To be honest, I wouldn’t be surprised by xAI working around this, but only because they’re not a serious competitor. OpenAI has their own models and doesn’t need to do this.
I get it’s expensive. But their whole goal is to automate AI research (capabilities or safety) and they claim that the “race” is so crucial to humanity that they be first. On top of all that, if they think this TOS phrase is sufficient deterrence to their competitors, then it makes me think they they either they’re lying about the stakes, or they’re incompetent.
I think the ToS is probably sufficient deterrence with the state of the race today, since it’s easier for their competitors to use their own models than to work around Anthropic blocks (and at least in the US there could be legal consequences to intentionally violating the ToS).
If they got sufficiently far ahead, they might not make their models public, and we already have a case of that with Mythos (although it’s unclear if they don’t want their competitors to have access, or if they just can’t afford to run it at scale).
Also even without the ToS, competitors may not want their engineers to leak all of their plans to Claude.
I would assume they would keep that version of Claude for themselves internally, and not the public version of Claude. Why give their competitors any edge at all?
I think the answer is that they don’t give their competitors access. The Terms of Service explicitly prohibit using Claude to build a competing model or service:
And they’ve cut off access for both OpenAI and xAI because of this.
There’s limits to how enforceable this is if users are motivated enough, but apparently they feel like this is sufficient.
The alternate strategy of making multiple frontier models would be very expensive.
I doubt lack of official access is a real issue for anyone motivated. Honestly, what keeps people from xAI or OpenAI from still using Claude Code? Getting personal access? Is Antrophic actively monitoring for anything OpenAI related and blocking those accounts?
The whole point of using an AI coding agent is to reduce the amount of effort involved in coding. If you have to jump through hoops to ensure that your usage never shows up on a company IP, is it actually worth the effort (especially since OpenAI has their own models)? Plus a lot of people won’t pay out of pocket just to help the company (even if it would obviously be worth it), and the company would have to consider legal risks to intentionally violating the ToS.
To be honest, I wouldn’t be surprised by xAI working around this, but only because they’re not a serious competitor. OpenAI has their own models and doesn’t need to do this.
I get it’s expensive. But their whole goal is to automate AI research (capabilities or safety) and they claim that the “race” is so crucial to humanity that they be first. On top of all that, if they think this TOS phrase is sufficient deterrence to their competitors, then it makes me think they they either they’re lying about the stakes, or they’re incompetent.
I think the ToS is probably sufficient deterrence with the state of the race today, since it’s easier for their competitors to use their own models than to work around Anthropic blocks (and at least in the US there could be legal consequences to intentionally violating the ToS).
If they got sufficiently far ahead, they might not make their models public, and we already have a case of that with Mythos (although it’s unclear if they don’t want their competitors to have access, or if they just can’t afford to run it at scale).
Also even without the ToS, competitors may not want their engineers to leak all of their plans to Claude.