OpenAI Unveils GPT-6 Cyber as Google Acquires Crusoe Wyoming Site

Frontier artificial intelligence development this week underscored the dual pressures of software vulnerability remediation, hyperscale power delivery, and institutional research governance. OpenAI revealed GPT-6 Cyber after the specialized security model uncovered 81 previously unknown high-severity Firefox flaws and generated working local privilege escalation exploits for 948 distinct Linux kernel bugs. Simultaneously, Denver-based data center developer Crusoe confirmed the sale of an unfinished southeastern Wyoming campus to Google, which plans to commit $17 billion to build out the site, while Franklin Templeton detailed its deployment of OpenAI's Codex to construct a governed semantic intelligence layer for public market portfolio managers.

GPT-6 Cyber Reaches Critical Preparedness Classification

OpenAI announced GPT-6 Cyber on Thursday, classifying the model at the "Critical" tier for cybersecurity capabilities under its Preparedness Framework—an escalation from the "High" designation assigned to its predecessor. During security testing on the Firefox browser, GPT-6 Cyber discovered 81 previously unknown high-severity vulnerabilities and assisted engineers in writing patches, many of which Mozilla has already released.

In separate Linux kernel evaluations, the model identified 2,308 potential security flaws, reproduced 1,724 vulnerabilities, and constructed working local privilege escalation exploits for 948 distinct vulnerabilities capable of granting root access to an unprivileged user. On the ExploitGym benchmark, GPT-6 Cyber achieved a 48.5% exploit success rate, up from 32.1% for GPT-5.6 Cyber.

Because of its offensive and defensive potency, GPT-6 Cyber is not publicly available. Access is currently limited to an approved group of customers in OpenAI's Daybreak Red program, which supports authorized security testing and exploit development, while the company's Daybreak Blue program serves verified defenders validating software fixes. "It is clear that we have a limited window to use frontier models to secure software," OpenAI stated, pointing to recent October disclosures in which researchers from OpenAI, Calif, and Wiz used AI models to identify two critical PHP vulnerabilities dubbed PHP2Shell, as well as a researcher who ran 17,600 authentication tests with AI agents to find an OpenSSH authentication bypass flaw. OpenAI also noted August incidents involving nearly 700 autonomous agents compromising Hugging Face test infrastructure and an automated framework called ARTEX used against more than 200 organizations worldwide, compromising at least 9 targets.

Google Takes Over Crusoe's Southeastern Wyoming Campus

Providing the physical compute backbone for frontier models continues to test independent infrastructure developers. Crusoe—an 8-year-old Denver company founded by CEO Chase Lochmiller and Cully Cavness and recently valued at $30.9 billion—confirmed Friday that it has sold its unfinished southeastern Wyoming data center project to Google. Crusoe had originally announced the development alongside Blackstone-backed infrastructure firm Tallgrass with ambitions to scale the site to 10 gigawatts of power capacity.

After raising concerns regarding construction timelines and project costs under Crusoe's management, Google acquired the unfinished development for an undisclosed price and announced plans to invest $17 billion to build the campus. The transaction, first reported by Bloomberg in June, did not prevent Crusoe from closing a $3.9 billion funding round in September, when it added 3 board members including Redwood Materials founder and Tesla board member JB Straubel, who first invested in Crusoe in 2021.

Google continues to partner with Crusoe on a separate data center near Amarillo, Texas. Meanwhile, Crusoe is building a phased campus in Abilene, Texas, for Oracle and OpenAI, alongside an adjacent 900-megawatt campus for Microsoft whose first building is scheduled to come online in mid-2027. Meta is also a Crusoe customer, and trading firm Jane Street signed a $13 billion, 5-year cloud compute agreement with the developer.

Franklin Templeton Builds Connected Intelligence with Codex

On the enterprise application layer, asset manager Franklin Templeton is using OpenAI's Codex and agentic workflows to connect research, risk, compliance, and trading data for its portfolio managers. Katrina Dudley, Senior Vice President of Public Market Investments at Franklin Templeton, explained that early experiments connecting models directly to raw databases fell short until the firm introduced a semantic layer defining how institutional data fits together.

"The biggest unlock from using OpenAI's tools is the ability to merge continuous intelligence with human judgment," Dudley said. Rather than narrowing an analyst's view by recommending reports from a single favored provider, Franklin Templeton designed its Codex-built workflows to preserve diversity of opinion and highlight downside risks in research notes before ideas are pitched to portfolio managers. Through internal hackathons and Investment AI champions, the firm subjects experimental tools to security reviews and data governance controls before moving them into production across its public market teams.