Google clamps down on OpenClaw users over Antigravity abuse

OpenClaw is an open-source autonomous agent built to run shell commands, interact with local files, and automate multi-step workflows. But on February 23, developers using it alongside Google's Antigravity vibe coding platform hit a hard wall.

Google restricted access for developers combining OpenClaw with Antigravity, citing malicious activity. Within hours, affected developers took to X and Y Combinator discussion boards, warning that running OpenClaw against Google services had cost them access to their entire Google accounts, including Gmail.

Siphoning tokens and crashing backends

According to Google, the issue came down to resource exploitation. Certain users were routing OpenClaw through Antigravity to siphon outsized batches of Gemini tokens, causing serious backend performance issues for legitimate Antigravity customers.

Varun Mohan, a Google DeepMind engineer and the founder of Windsurf, addressed the backlash on X. He stated that the team saw a massive surge in malicious usage that degraded platform stability, forcing them to quickly shut off unapproved access methods.

Mohan acknowledged that some users may not have realized they were violating the platform's Terms of Service, adding that Google is looking into pathways to restore their access while balancing limited server capacity. A Google DeepMind spokesperson also confirmed to VentureBeat that the crackdown aims to bring usage into compliance with service terms, rather than permanently ban third-party platforms.

A messy timing and a severed bridge

The timing of the crackdown raised plenty of eyebrows across the industry. Just one week earlier, on February 15, OpenAI CEO Sam Altman announced that OpenClaw creator Peter Steinberger had joined OpenAI to direct its next generation of personal agents.

Even though OpenClaw technically remains open-source under an independent foundation, it is now strategically and financially tied to Google's chief rival. By cutting off OpenClaw, Google effectively severed an easy pipeline allowing an OpenAI-aligned tool to tap into its frontier Gemini models.

In response, Steinberger announced that OpenClaw would completely drop support for Google products, escalating the feud. The clash echoes Anthropic's moves over the past year, when it throttled Claude Code over continuous 24/7 misuse and later rolled out client fingerprinting to block wrappers like OpenClaw from tapping its models.

Why it matters

For builders and power users, the immediate takeaway is operational risk. Hooking up an unvetted autonomous agent to your primary Google ecosystem or personal Gmail is a dangerous gamble. If the provider flags the traffic as abusive, you risk losing your core identity along with your API access.

For the broader AI ecosystem, this marks the slow death of the "bring your own agent" era. Frontier model labs are moving fast toward closed, vertically integrated gardens. They want full telemetry and direct subscription dollars, leaving little room for open-source wrappers that run up massive compute bills.

We think Google had every right to protect its compute capacity from being drained by an OpenAI-adjacent project. However, bricking developer accounts down to their personal Gmail feels excessively punitive and risks driving enthusiastic power users straight into rival ecosystems.