OpenAI's GPT-6 Astra First to Hit “Critical” Cyber Bar, Worldcoin (WLD) in Focus

Guards at OpenAI Offices Vote to Strike

Security guards who protect the offices of OpenAI, Anthropic, Google and Salesforce voted this week to authorize a strike, pushing a wage dispute to the brink of direct action at the companies at the heart of the AI boom. The workers, organized under SEIU-United Service Workers West, are employed through major contractors Allied Universal and Securitas, and the vote clears the way for a walkout if stalled talks collapse. The union is demanding a minimum hourly wage of $30, up from the current $22, along with expanded benefits. Progress has been minimal: after 19 bargaining sessions, Allied Universal has offered only a 25-cent-per-hour raise set to take effect in 2027, with no further increases proposed over the following three years. Union president David Huerta described the offer as shameful, wrong and deeply disrespectful, and guards demonstrated in San Francisco on Wednesday against the stagnant negotiations.

The dispute matters because security has become a core operating cost for AI firms. As anti-AI sentiment hardens, some employees and executives have faced threats and even violence. An analysis by The Wall Street Journal found that Palantir, Oracle and Salesforce all sharply increased security spending in 2025, and San Francisco police have received threat reports involving OpenAI and Anthropic staff, with some firms deploying armed guards. Last week Anthropic asked some employees to work remotely in anticipation of a possible walkout — a step Huerta said shows the company understands the guards’ importance even if the contractors bargaining with them do not. That labor pressure around AI campuses is a relevant signal for the wider ecosystem, including Worldcoin (WLD), the identity project built by Altman-affiliated Web3 firm Tools for Humanity.

GPT-6 Astra Crosses the Critical Threshold

Hours after that labor story broke, OpenAI delivered the sector’s bigger headline: the release of GPT-6 Astra. Launched in the early hours of September 4, the model targets computer operation, software engineering, cybersecurity, scientific research and professional work. OpenAI’s official announcement confirms Astra scored 57.9% on Terminal-Bench 4.0, well above the 37.3% of GPT-5.6 Sol, and reached 100% on ExploitBench — the first OpenAI model to cross the Critical cybersecurity capability threshold under its Preparedness Framework, a designation for models that could materially aid real-world attacks. During testing, the model also discovered and exploited two previously unknown zero-day vulnerabilities, extending what we noted when GPT-6 Astra posted 98.6% on ARC-AGI-3 and again on its 100% ExploitBench score. Access starts with a small group of institutions, then reaches ChatGPT Plus, Pro, Business and Enterprise users over the coming days, alongside the OpenAI API, Microsoft Azure and AWS Bedrock. API pricing is set at $10 per million input tokens and $50 per million output tokens. Capital is following the capability jump: a report citing people familiar with the matter says quant trading firm Jane Street signed a five-year, roughly $13 billion cloud contract with AI provider Crusoe, averaging about $2.6 billion a year in committed spend for GPU clusters used in training and inference. Jane Street already runs tens of thousands of high-end GPUs and builds models that serve its strategies directly — evidence that large-scale compute buyers now extend beyond frontier labs into financial institutions operating what amounts to an industrial-scale AI trading bot. Neither side has disclosed the GPU count or deployment timeline. Readers tracking the market in real time can follow live spot and futures prices on MEXC.

WLD’s Altman Link Back in Play

The two developments share one arc: the AI economy is expanding simultaneously into security labor, industrial compute and headline model capability, and markets have repeatedly read that momentum as a proxy signal for the Worldcoin ecosystem. WLD, issued by Altman-linked Tools for Humanity, has historically traded as an AI-theme vehicle among altcoin assets; our desk notes spot slipped 3.6% over the past 24 hours, a move that looks like broad risk-off rather than a reaction to either story. OpenAI’s announcement confirms the Critical designation, the benchmark results and the API pricing, but does not name the vendors affected by the two zero-days or give a disclosure timeline — an omission worth tracking as enterprise adoption widens. If Astra’s capability jump sustains the AI narrative, market cap rotation into AI-linked crypto names, as seen when OpenAI agents left 18,000 posts on German wiki, could reassert — keeping Worldcoin in focus.