ChatGPT sign-in and subscriptions land on OpenClaw

Sam Altman just turned OpenAI’s $23 subscription into the official backend for the hottest open-source project in the world, directly countering Anthropic’s clampdown.

At 2:33 a.m. on May 2, Altman casually posted on X that ChatGPT subscribers can now sign in to OpenClaw via OAuth and run autonomous AI agents powered by GPT-5.4 without paying extra API fees. "happy lobstering," Altman added, signaling a massive land grab in the AI agent space.

The runaway lobster

OpenClaw started in November 2025 when Austrian developer Peter Steinberger built a local agent tool in a Madrid cafe, originally naming it Clawdbot with a lobster logo. Anthropic sent a trademark complaint, prompting a quick rebrand to Moltbot and eventually OpenClaw, though the lobster stayed.

The tool lets users run autonomous agents locally across messaging apps like WhatsApp, Telegram, Signal, Discord, Slack, iMessage, and Teams. It handles email, manages calendars, writes code, and executes multi-step tasks while keeping data on user hardware. It exploded, crossing 346,000 GitHub stars and 3.2 million users in under five months. Jensen Huang called it the most popular open-source project in history at Nvidia’s GTC in March, after it beat React’s ten-year star record in just 60 days.

Altman acted quickly. In February, he hired Steinberger to lead next-gen personal agents at OpenAI and moved OpenClaw into an independent foundation with OpenAI funding. Sequoia even handed out 200 engraved Mac Minis at an event, confirming OpenClaw as the agent layer everyone wanted to integrate with.

Opposite bets on agent compute

Anthropic and OpenAI are now running completely different playbooks.

On April 4, Anthropic cut off Claude Pro and Max subscribers from using their flat-rate plans with OpenClaw. Autonomous agents trigger thousands of API calls a day, and Anthropic decided that unlimited background compute was torching its margins.

OpenAI did the exact opposite. By allowing ChatGPT Plus members to run GPT-5.4 via the Codex endpoint inside OpenClaw, OpenAI is absorbing the compute surge to lock down distribution. If a fraction of OpenClaw's 3.2 million users convert to paying $23 a month for ChatGPT Plus, OpenAI gets an entrenched developer audience no ad campaign could buy.

Security, however, remains shaky. In January, researchers uncovered a severe remote code execution flaw (CVE-2026-25253) that let malicious websites hijack local agent instances over unvalidated WebSockets. A recent audit of ClawHub, its community skills marketplace, also identified 824 malicious skills out of 10,700, and over 30,000 OpenClaw instances were found exposed to the open internet.

Why it matters

For power users and developers, this makes running complex personal agents drastically cheaper. Instead of racking up massive pay-as-you-go API bills, you can run a 24/7 background agent on GPT-5.4 for a predictable $23 monthly fee. That said, everyday users should tread carefully: giving an open-source agent full disk access and messaging permissions while thousands of malicious skills float around the marketplace is a massive security risk.

Strategically, OpenAI's move is brilliant. Rather than attempting to build a competing agent client from scratch, it simply adopted the dominant open-source ecosystem and turned Anthropic's defensive retreat into an offensive distribution victory.

We think the big question now is sustainability. Autonomous background loops generate massive inference costs compared to human chat prompts. OpenAI is effectively subsidizing heavy compute to monopolize agent mindshare, but if millions of agents keep hammering Codex around the clock, rate limits or pricing tiers will almost certainly follow.