Why Anthropic’s Stance on Open Source Has Lost My Support

When Anthropic emerged from OpenAI, the narrative was undeniably compelling. Here was a team of researchers who claimed to care about AI safety, transparency, and alignment above all else. They promised a more thoughtful approach to building artificial intelligence, positioning themselves as the responsible adults in a room full of move-fast-and-break-things hype.

Years later, that polished facade has worn off. What remains is a corporate entity that uses safety rhetoric as a marketing tool, delivers a product experience burdened by opaque restrictions, and actively lobbies against the democratisation of artificial intelligence through open-weight models.

Pulling Up the Ladder on Open Weights

The most concerning aspect of Anthropic’s trajectory is its ideological and political hostility towards open-weight models. Under the banner of risk mitigation and planetary safety, Anthropic has consistently supported regulatory frameworks that place disproportionate burdens on open-source developers.

The argument they advance is simple: advanced AI models are too dangerous to be released into the wild, so access must be restricted to controlled, proprietary APIs.

This position is both intellectually dishonest and deeply self-serving. The entire modern technology ecosystem, including the cloud infrastructure, training pipelines, and core algorithmic frameworks that Anthropic relies upon, was built on open research, open-source software, and public datasets. To leverage that collective human effort to build a proprietary system, and then immediately turn around to lobby for regulations that render open distribution illegal or practically impossible, is classic ladder-pulling.

Open-weight models like Meta’s Llama series, Mistral, and community-driven fine-tunes are essential for digital sovereignty. They allow researchers, small businesses, and individual developers to run systems locally without relying on a centralised corporate proxy. They permit auditing, local fine-tuning, and long-term stability without the risk of an API provider arbitrarily changing terms, raising prices, or deprecating a model overnight.

Framing open weights as an inherent security threat serves a singular commercial purpose: creating regulatory capture. By convincing lawmakers that only a handful of well-funded corporations can safely host frontier models, Anthropic isn't protecting humanity; it is securing a legal moat around its business model.

Paternalistic Marketing and Moral Superiority

Anthropic’s public relations strategy relies heavily on a high-minded, paternalistic tone. Where competitors sell speed, integration, or raw capability, Anthropic sells moral rectitude. Terms like "Constitutional AI" and endless position papers on extreme existential risks are marketed not just as technical achievements, but as proof of superior corporate ethics.

This approach feels increasingly hypocritical when contrasted with their financial backers. Anthropic has raised billions from hyperscalers like Amazon and Google. They are deeply embedded in the traditional venture capital ecosystem, subject to the exact same pressures for exponential return on investment as every other enterprise in Silicon Valley.

When a company funded by the world's largest data monopolies claims to be acting purely in the public interest while pushing to restrict public access to code, the narrative breaks down. The safety-first messaging is no longer a pledge of accountability; it is a slick, corporate marketing campaign designed to differentiate a closed product in a crowded market.

A Frustrating Product Reality

This top-down, protective ethos directly degrades the actual user experience of building with their models.

In practice, working with Claude often feels like interacting with a system designed by a legal compliance department rather than an engineering team. Benign technical queries, creative code refactoring, and complex analytical prompts regularly trigger heavy-handed system refusals or patronising lectures on safety.

Furthermore, product delivery from Anthropic suffers from persistent operational inconsistencies:

Opaque rate limits: Sudden enforcement of aggressive usage caps on paid tiers, often without clear metrics or advance notice.
Silent guardrail adjustments: Changes to backend system prompts that alter model behaviour, tone, and capability mid-project.
Refusal false-positives: Overly sensitive safety filters that halt legitimate software development, vulnerability research, and systems administration tasks.

For developers attempting to build reliable, production-grade applications, this environment is deeply frustrating. A tool must be predictable, transparent, and controllable. Anthropic's insistence on maintaining an intrusive layer of corporate oversight between the user and the output makes their platforms fundamentally unreliable for long-term integration.

Moving Beyond Closed Gates

The trajectory of technology has repeatedly proven that centralised, closed ecosystems ultimately stifle innovation. The future of software, including artificial intelligence, belongs to open standards, transparent weights, and decentralised infrastructure.

Anthropic’s attempt to position closed-source, API-gated models as the only safe way forward is a regressive vision for computing. By prioritising regulatory capture, hiding behind virtue-signalling marketing, and delivering an API experience constrained by paranoia, they have alienated the very community of builders who originally rooted for their success.

True safety in technology does not come from corporate gatekeepers or closed APIs. It comes from open inspection, broad distribution, and giving individuals the freedom to own, run, and understand the tools they use.

I'll stick with the others, thanks.