AI + Careers Briefing — September 25, 2026

Three AI-career stories crossed the same day. New CNBC/Handshake data spells out what a resume actually has to prove now; a federal appeals court upheld one of two Pentagon "risk" labels on Anthropic (the other's been ruled unlawful), reshaping the DoD-adjacent AI job market; and Microsoft merged Copilot's chat, coding and agents into one app, chasing OpenAI's and Anthropic's enterprise push. None of it forces a change of plan today. Together they price the same skill: bet on AI-adjacent capability you can point at, not on the labels of the tools you've touched.

What should recent grads actually change on their resume for an AI-driven job market?

Tailor each resume to the specific listing, and lead with AI skills backed by a real project instead of a generic skills-list line. That's the practical advice in a CNBC roundup published September 25 on how college students and recent graduates should approach AI, resumes and the job search, set against a market that looks tougher at the entry level. The unemployment rate for college graduates aged 22 to 27 hit 5.7% in June, a full percentage point higher than it was two years earlier, CNBC reported. Among graduating seniors, 74% of AI-skill mentions on resumes are now tied to hands-on projects rather than classroom coursework, CNBC found — a shift that lines up with employer demand: Handshake's 2026 workforce outlook shows the share of full-time postings on its platform mentioning generative AI up almost 5 times since 2023, with internship postings up more than 4 times.

The takeaway isn't adding an AI-skills bullet — it's proof. A resume claim needs a specific tool, task and result, the same real-project detail that CNBC's reporting says now separates credible mentions from buzzwords. Treat tailoring as non-negotiable, not optional: rewriting each application with an ATS-aware pass matters more against an entry-level pool where AI-related postings are growing several times faster than hiring overall. The skills worth prioritizing are the ones tied to an outcome you can describe in an interview, not the ones easiest to list.

Source: CNBC; Handshake

Did an appeals court just uphold the Pentagon's "risk" label on Anthropic — and what does it mean for AI job seekers?

Yes, but only one of two. A divided panel of the U.S. Court of Appeals for the D.C. Circuit ruled 2-1 on Friday that the Pentagon's designation of Anthropic as a "supply chain risk" can stand, according to CNBC. The designation the court upheld is one of two the Defense Department has used against the AI company, forcing Anthropic to fight the label in two separate courts.

The fight traces back to a $200 million Pentagon contract Anthropic signed in 2025 for its Claude models. Talks over deploying Claude on the Defense Department's GenAI.mil platform collapsed after Anthropic refused to guarantee the military unrestricted use of its technology, insisting on safeguards against fully autonomous weapons and domestic mass surveillance. The Pentagon responded by declaring Anthropic a risk to the federal supply chain. That split matters: a federal judge in San Francisco ruled last month that the companion designation was unlawful — amounting to First Amendment retaliation and a due-process violation — and wrote that "the empty invocation of national security is not a blank check to punish and retaliate against government critics," TechCrunch reported. Friday's D.C. Circuit ruling, by contrast, upheld the other designation: the majority concluded Claude's built-in restrictions and the contract dispute made the model unreliable for military use, while dissenting Judge Karen LeCraft Henderson argued the law doesn't treat honest enforcement of a company's own restrictions as a supply chain risk, per ABC News. An Anthropic spokesperson told ABC News, "Another federal court has already held the government's parallel designation unlawful. We remain confident in our position and are considering all options, including further review."

For anyone chasing a role at an AI lab with federal ambitions, the split rulings signal real hiring risk on both sides. Companies flagged as a supply-chain risk lose eligibility for federal contracts overnight, which can freeze growth plans for sales and solutions-engineering teams built around government work. At the same time, labs caught in designation fights are staffing up compliance, trust-and-safety, and government-relations positions to manage exactly this kind of dispute — work that sits between legal, policy, and engineering rather than pure model-building. Job seekers eyeing DoD-adjacent AI work should check a company's active designations and litigation status before applying, since a single appellate ruling can add or erase openings overnight.

Source: CNBC; ABC News; TechCrunch

Why is Microsoft merging all its business AI into one Copilot app — and which skill does that argue for?

Microsoft rolled out a redesigned Copilot app for business customers on Friday that merges chat, coding and autonomous AI agents into a single interface, a direct response to enterprise customers drifting toward bundled offerings from Anthropic and OpenAI. The app is organized into three sections: Home, which pairs chat with Microsoft's Cowork assistant; Code, which lets non-developers build apps and dashboards by describing them in plain language; and Autopilot, where users assign an agent a name, a role and a goal and let it run continuously on tasks like tracking project deadlines, according to GeekWire and Fortune.

Microsoft framed the redesign as a competitive answer to rivals pushing into enterprise. OpenAI launched its own consolidated "super app" in July, per Fortune. Copilot executive vice president Jacob Andreou took a direct jab at a rival approach, saying that "sending your most sensitive files down to a local machine into a VM was a complete nonstarter," according to GeekWire. The adoption numbers explain the urgency: Microsoft said paid Microsoft 365 Copilot seats surpassed 30 million in July (per Fortune), up from 20 million in April (per GeekWire) — but GeekWire noted that still represents only about 7 percent of the more than 450 million commercial Microsoft 365 seats the company reported in January. Rather than charging a flat subscription for the new capabilities, Microsoft is shifting the Code and Autopilot portions of the app to usage-based billing, per Fortune.

Copilot's paid-seat adoption still sits in single digits, so most workers won't be required to master Microsoft's specific tool this fall. What matters more is the skill Microsoft is betting on: directing autonomous agents rather than just prompting a chatbot. Job postings that mention "agentic AI," Copilot, or Claude are increasingly testing for candidates who can assign a goal to an AI system, check its output and course-correct — not just write a good prompt. Because Microsoft, OpenAI and Anthropic are all racing toward the same agent-based model, that skill is becoming vendor-agnostic. Job seekers who can point to hands-on time managing an AI agent's task queue, even in a personal project, will read as more current than candidates who only list "ChatGPT experience" on a resume. See our guide to AI skills tech employers are hiring for for the specific competencies showing up in listings.

Source: GeekWire; Fortune

What to watch

• The San Francisco district-court ruling on the companion Anthropic designation heading toward appellate review — if that ruling is upheld too, the same law will have produced conflicting outcomes on two designations against the same company, teeing up a Supreme Court moment for federal AI-vendor policy.
• Whether Microsoft's 30-million-seat number moves meaningfully above ~7% penetration in the next quarterly update; that's the leading indicator for how quickly "agentic AI" skills clear the hiring bar broadly, not just at Microsoft-shop employers.
• Handshake's next Class of 2027 update — whether the 5x AI-mention growth trajectory continues, or plateaus as employer wording standardizes.