The race to automate desktop workflows just got real—and risky. A new hands-on comparison of OpenAI's ChatGPT Work and Anthropic's Claude Cowork reveals both tools deliver similar automation capabilities, but only one currently raises serious safety concerns for enterprise users. According to testing published by ZDNet, the performance gap is narrow, but the trust gap is wide—and that could reshape which AI assistant wins over corporate IT departments.
OpenAI and Anthropic are locked in a high-stakes battle for desktop dominance, and the first real-world comparison between their flagship automation tools reveals an uncomfortable truth: similar capabilities don't guarantee equal trust.
ChatGPT Work and Claude Cowork represent the next frontier in AI-powered productivity—tools that don't just answer questions but actively manipulate files, navigate applications, and execute complex workflows on behalf of users. According to hands-on testing detailed by ZDNet, both products deliver on their automation promises with surprisingly similar results. But when it comes to the critical question of safety, only one assistant left the tester feeling comfortable letting it run unsupervised.
The testing reveals that Claude Cowork currently feels 'considerably safer' than ChatGPT Work, despite both tools wielding similar levels of system access. That's not a minor distinction—it's potentially a dealbreaker for enterprise IT teams already nervous about granting AI agents deep permissions into corporate file systems and applications. While the report doesn't specify the exact security mechanisms that differentiate the two, the perception gap alone matters in a market where trust is currency.
Both OpenAI and Anthropic have bet big on workplace automation. OpenAI's ChatGPT Work builds on the company's massive consumer success with ChatGPT, extending those capabilities into desktop environments where the stakes are infinitely higher. A misstep with a personal query is annoying; a mistake with corporate financial data could be catastrophic. Anthropic, meanwhile, has positioned Claude as the safety-conscious alternative since its founding by former OpenAI researchers, and Claude Cowork appears to be living up to that reputation.
The similar performance between the tools suggests both companies have cracked the technical challenge of desktop automation. Users can expect comparable results whether they're asking ChatGPT Work or Claude Cowork to organize files, extract data, or automate repetitive tasks. That parity shifts the competitive battleground from raw capability to guardrails, transparency, and user control—areas where enterprise buyers are increasingly demanding proof, not promises.
Timing matters here. This comparison arrives as AI agents transition from research projects to mission-critical business tools. Companies like Microsoft are embedding similar capabilities into Office 365 through Copilot, while startups are racing to build specialized automation layers for everything from customer support to financial analysis. The question isn't whether AI will automate desktop workflows—it's which AI companies will earn the privilege.
The 'nervous' reaction to ChatGPT Work described in the testing is particularly striking given OpenAI's market leadership and massive enterprise adoption of ChatGPT Enterprise. It suggests that consumer AI success doesn't automatically translate to workplace trust, especially when tools gain permission to read, modify, and delete user files. Anthropic's apparent safety advantage could be architectural—better permission models, clearer user confirmation flows, or more conservative default behaviors—but the report's lack of specifics leaves room for interpretation.
What's clear is that both tools share similar strengths, likely including natural language understanding, multi-step task execution, and integration with common desktop applications. The automation use cases are straightforward: batch file processing, data extraction from documents, cross-application workflows, and repetitive administrative tasks. For knowledge workers drowning in busywork, either tool could deliver meaningful time savings.
But productivity gains mean nothing if users don't trust the tool enough to actually use it. Enterprise software history is littered with technically impressive products that failed because they couldn't overcome user anxiety. If ChatGPT Work consistently makes users nervous while Claude Cowork doesn't, that's an adoption barrier OpenAI will need to address fast—especially as Anthropic continues positioning itself as the responsible AI alternative.
The competitive dynamics extend beyond just these two players. Google is developing similar capabilities through Gemini, Microsoft has Copilot deeply integrated into Windows, and Apple is expected to unveil its own desktop AI features. The company that figures out how to deliver powerful automation without triggering user anxiety will have a massive advantage as AI assistants become standard workplace tools.
The desktop automation wars are heating up, and technical capability alone won't determine the winner. OpenAI's ChatGPT Work and Anthropic's Claude Cowork may deliver similar results, but the safety perception gap revealed in this testing could prove decisive as enterprises decide which AI assistant gets access to their most sensitive workflows. For OpenAI, the challenge is clear: match Claude's comfort factor or risk losing enterprise buyers who prioritize trust over speed. For Anthropic, the opportunity is equally obvious—lean into that safety advantage while it lasts. As these tools evolve from experiments to everyday essentials, the company that makes users feel secure will likely capture the larger share of the workplace automation market.
