The Omniscient Bulletin · 2026-06-23
The Omniscient Bulletin — June 23, 2026
AI moves off the demo stage and into the filings, the payrolls, and the power plants
No new flagship model led Monday, and it did not need to. AI showed up instead where it now matters more: in a regulatory filing, on 125,000 desks, and in a 20-year power contract. Oracle told the SEC that deploying AI had helped shrink its headcount by 21,000; hours later Samsung handed ChatGPT and Codex to roughly 125,000 employees. The same technology thinned one payroll and was wired into another. The rest of the day filled in the bill: gigawatts of gas-fired power, a $4 billion bet against Nvidia's software stack, and a lab pushing its cyber model out to anyone who can prove they defend.
Frontier
OpenAI expands its cyber model to verified defenders, taking a more open path than Anthropic
OpenAI released an updated version of GPT-5.5-Cyber, its model tuned for red-teaming and penetration testing, through a continued limited release to trusted defenders. It also shipped a Codex Security update that moves from finding bugs to landing fixes, launched a Daybreak Cyber Partner Program to embed the model inside commercial security products, and announced Patch the Planet to fund hardening of critical open-source code. The approach contrasts with Anthropic, which kept its strongest cyber model behind tight vetting before Washington forced it offline.
Labor
Oracle cuts about 21,000 jobs and names AI as a cause in its annual filing
In its annual report, Oracle disclosed that its workforce shrank by about 21,000 people, or 13 percent, over the past year, to roughly 141,000 as of May 31. The filing states that 'the adoption and deployment of AI technologies across our operations have resulted, and may continue to result, in reductions to our workforce,' and restructuring costs reached about $1.8 billion. It is among the clearest cases yet of a profitable megacap naming AI, in a regulatory filing, as a cause of large-scale job cuts, even as it pours tens of billions into AI data centers.
Industry
Samsung hands ChatGPT and Codex to about 125,000 employees
Samsung Electronics deployed OpenAI's ChatGPT Enterprise and the Codex coding agent to its entire South Korean workforce, about 125,000 people, plus all staff in its global Device eXperience division. OpenAI called it one of its largest enterprise rollouts ever, putting Codex in front of non-developers across software, product, manufacturing, and marketing. The move reverses a 2023 internal ban on generative AI, and it deepens a relationship in which Samsung also supplies the memory chips that OpenAI's data centers run on.
Compute
Chevron will build a 2.7-gigawatt gas plant to power a Microsoft AI data center
Chevron signed a 20-year deal to sell Microsoft power from a new gas-fired complex it will build in Reeves County, West Texas, one of the largest gas-powered data center projects in the country. The site spans more than 2,000 acres and is designed to ramp to about 2.67 gigawatts, with electricity flowing by 2028 from GE Vernova turbines fed by Permian Basin gas. Chevron projects more than $10 billion in tax revenue and roughly 2,000 jobs. An oil major building dedicated power for AI shows how generation, not the grid, has become the binding constraint on the buildout.
Compute
Qualcomm is reportedly nearing a $4 billion deal for AI-software maker Modular
Qualcomm is in advanced talks to buy Modular, the AI software startup founded by Swift and LLVM creator Chris Lattner, for about $4 billion, according to Bloomberg. Modular builds the compiler and runtime layer that lets models run across different chips without rewriting code, a direct challenge to the lock-in behind Nvidia's CUDA. The price is roughly 2.5 times Modular's $1.6 billion valuation from nine months ago, and it follows Qualcomm's reported pursuit of chipmaker Tenstorrent. The deals show Qualcomm assembling a full silicon-and-software stack to contest the data center.
Policy
The Army picks Anduril and Palantir to build the data layer of its top command system
The Army selected Anduril to lead the common data layer for Next Generation Command and Control, its highest-priority modernization program, in the first major award as the effort moves past prototyping. Anduril's Lattice will serve as the tactical data layer and Palantir's Foundry as the cloud layer, two pieces of a planned four-layer stack. The work falls under Anduril's 10-year enterprise agreement with the Army, which has a ceiling of up to $20 billion; no figure was given for this award. It puts two AI-native firms, not legacy primes, at the core of the service's battlefield network.