In practice, Cowork may therefore not be viable for South African organisations handling personal information until Anthropic offers regional data storage or stronger compliance tooling around this product. (Image created by ChatGPT). Anthropic launched a legal plugin for Claude Cowork at the end of January, and the reaction was immediate. Stock prices in legal publishing and software companies tumbled. Bloomberg ran headlines about sinking valuations. Commentary ranged from measured curiosity to something closer to existential panic. The reality, for legal functions across organisations, whether law firms, in-house legal departments, compliance teams, contract management functions, or the legal AI providers serving them, is less dramatic.
The plugin does some genuinely useful things. It can review contracts clause by clause against a configured negotiation playbook, triage non-disclosure agreements (NDAs), flag compliance issues, and generate templated responses for common legal enquiries. It is configurable to an organisation's risk tolerances. That is useful, but not the paradigm shift some coverage implies.
What Anthropic has done is sensible product development. It has taken its powerful general-purpose models and done the engineering work to make it more immediately useful for specific industries. The legal plugin bundles pre-built tools and workflows so that Claude behaves more like a domain specialist than a very clever generalist. Anthropic has done for its legal users what those users would otherwise need to do themselves or pay someone to do for them. The right frame is not Anthropic suddenly competing head-to-head with the dedicated legal AI providers. It is Anthropic making Claude more accessible and more immediately productive for people working in legal, and in a variety of other industries for which they have developed plugins.
But wait, there's still a gap
The more interesting question is about the gap that remains, and it is a gap worth understanding.
Most legal professionals are not tech-native. They are not going to write prompts from scratch, configure playbooks in a terminal, or integrate MCP servers into their workflows. Legal functions across industries have made massive strides in AI adoption, but there is a material difference between using a purpose-built tool and extracting value from an agentic platform that expects technical fluency. That difference is not trivial, and it is precisely where companies like Harvey, Legora, Vincent, and others have invested most heavily.
These providers have put serious engineering effort into the layers around the model: document retrieval, structured workflows, database integrations, no-code customisation, enterprise security, and onboarding programmes that drive actual adoption. These are not lightweight products. The Cowork plugins do some of that, but at their core remains a set of open-source prompts and configurations on GitHub (as a brief aside, this makes them somewhat transferable to any of the AI tools an organisation might use). For a tech-savvy lawyer or an innovation team with development resources, it is a powerful starting point. For the average legal professional without dedicated technical support, it is not a replacement for a dedicated platform.
The dedicated legal AI providers have built their products for a particular sweet spot. They have targeted legal professionals who are not deeply technical but who are savvy enough to navigate a well-designed tool. They have targeted legal functions and departments that want to deploy these platforms with relative ease, organisations that might configure some workflows and build out some databases but broadly expect the product to work as it arrives. That has been the right strategy for the stage of adoption the industry has been at. But that very success has created a new dynamic: as lawyers have become more comfortable with AI tools and seen their potential, they have developed an appetite for deeper integration and greater customisation.
Who are these tools best suited for?
However, the market is maturing, in no small part because of what the dedicated legal AI providers have achieved. The organisations deploying these tools most aggressively, large law firms, corporates, financial institutions, and technology companies, typically have dedicated innovation and technology teams with the resources to customise extensively. Their legal, compliance, and contract management functions increasingly expect the same level of integration available to other business units.
These organisations often have substantial internal development capabilities already serving other business functions. For them, deploying Claude Cowork with different plugins configured for different teams may prove more appealing than procuring a separate legal AI platform for the legal department alone. The legal AI specialists have, in effect, educated these organisations about what is possible, and in doing so have raised expectations beyond what a standardised platform can satisfy. These organisations increasingly want to integrate their own infrastructure, build bespoke workflows reflecting specific practice areas, and exercise the kind of deep configurability that a general-purpose platform makes possible.
Potential red flags for SA companies
What are the practical realities of deploying Claude Cowork with the legal plugin?
For South African organisations considering implementation, the threshold question is what data you are processing and where the data goes. Anthropic’s Claude for Work plans currently store data at rest in the United States, with no regional alternative. Cowork, the desktop tool where the legal plugin runs, also processes data on US servers and is not yet captured in Enterprise audit logs or compliance exports. This raises cross-border transfer questions under POPIA, which requires that the receiving country offer data protection substantially similar to South Africa’s. Given the parallels with the GDPR and the 2020 Schrems II finding that US protections were inadequate, organisations processing data subject to POPIA would need to carefully assess whether consent or contractual mechanisms are sufficient.
In practice, Cowork may therefore not be viable for South African organisations handling personal information until Anthropic offers regional data storage or stronger compliance tooling around this product.
To ensure data protection compliance, the alternative is accessing Claude's capabilities via API through major cloud providers that have data centres in regions more straightforward from a POPIA perspective. Because Cowork legal plugin is not available through these channels, an organisation would need to build its own application layer to mimic Cowork, including the interface for lawyers, document management integration, and the logic that drives legal review.
The total cost of ownership
Where personal information is not being processed or is less of a concern, Anthropic offers Claude for enterprise use through various plans. Its Team plan offers simpler per-seat pricing with basic enterprise features, while the Enterprise plan adds audit logging, advanced security controls, custom data retention, and tailored terms. For organisations subject to rigorous compliance and regulatory standards, the Enterprise plan's capabilities are likely essential, though costs vary and can escalate significantly depending on usage.
Beyond the fees payable to Anthropic for Claude access, implementation of Cowork requires meaningful internal investment. The legal plugin's underlying configurations are publicly available, but translating them into workflows that lawyers actually use requires more than just making the tool available. At minimum, organisations will need to configure playbooks reflecting their specific risk tolerances, integrate document management systems, establish data governance protocols, build internal guidance on appropriate use cases, and invest in employee upskilling to drive adoption. The total cost of ownership extends well beyond the per-seat licensing fee, and for some this means the slightly higher per-seat licence fee of the legal AI providers may make more sense.
General-purpose AI providers will likely keep pushing into verticals. Anthropic, in particular, appears to be orienting itself firmly towards enterprise use cases. Cowork, the legal plugin, and the MCP protocol they developed and open-sourced, all signal a company that wants its model embedded in professional workflows. OpenAI, until very recently, seemed more focused on the consumer side, but this is changing.
In the other direction, the dedicated legal AI providers will need to open up. They will need to make their platforms more customisable and more accommodating of organisations that want to shape the tool around their own processes, priorities, and technology stacks. The platform-as-delivered model will continue to serve a large portion of the market, but it will not satisfy organisations at the top end that have both the ambition and the resources to build something distinctly their own.
And so, we get a narrowing gap: general-purpose providers adding vertical tooling, edging closer to what the specialists offer; specialists opening their platforms, edging closer to the flexibility the general providers afford. The irony is that the specialists' success in demonstrating the value of legal AI has accelerated demand for the very customisation that general-purpose platforms are well-positioned to provide. The durable competitive advantage for legal AI providers was never the model-which they do not own-but rather the application layer development they have built around the models to make them more usable for lawyers, the enterprise security, ease of deployment, and the strength of enterprise relationships built over years of embedding themselves in how legal functions across industries actually work.
The Claude Cowork legal plugin is not the disruption that sent share prices tumbling. But it is a clear signal about where the market is heading. The legal AI providers that read it correctly will still be leading this space in three years. The ones that dismiss it may find that convergence arrived faster than expected. DM
Aalia Manie is partner and director at Webber Wentzel Fusion; and Tristan Marot is an innovation lawyer at Webber Wentzel Fusion. Webber Wentzel Fusion is the applied AI and legal operations subsidiary of Webber Wentzel.