AI Security Gaps, Outsourcing Shifts.
A critical flaw in Cursor AI exposes firms using AI coding tools. Plus: what Cognizant's $600M acquisition signals for outsourced finance work.
A Critical Flaw in Cursor AI Lets Hackers Execute Code — Your AI-Assisted Workflows Are a Target Too
Researchers disclosed CVE-2026-26268 this week: a high-severity vulnerability in Cursor AI that lets attackers run malicious code the moment a developer clones a repository with hidden Git hooks. This isn't a niche developer problem. If your firm is using any AI-assisted tooling built or maintained by engineers using Cursor — and a lot of boutique fintech and AP automation vendors are — your pipeline could be touching compromised code without anyone knowing. Vendor security questionnaires need to start asking specifically about the AI IDEs their developers use.
The AI your vendor's developer used to write your automation is now part of your attack surface.The Accountants Who Will Be Hardest to Replace Are the Ones Who Can Audit AI Output
Cognizant just paid $600 million to acquire Astreya and bulk up its AI infrastructure services — that's a direct signal that large outsourcers are betting on owning the pipes, not just the labor. The accountants who survive that consolidation aren't the ones doing the work faster; they're the ones who can look at an AI-generated reconciliation or close package and know exactly where it's likely to be wrong. That skill looks like skepticism applied systematically: knowing which transaction types trip up ML models, where training data goes stale, and what a hallucinated journal entry actually looks like in a ledger. This week, pick one output from your AI tooling and manually trace it back to source. Do it like you're reviewing junior staff work.
Learn to review AI output the same way you'd review a first-year associate's work — with specific, documented checkpoints.Cognizant's Astreya Acquisition Is a Preview of Who Will Own AI Infrastructure for Mid-Market Finance
Cognizant acquiring Astreya for ~$600M is less about IT staffing and more about owning the AI infrastructure layer that sits underneath outsourced finance functions. For mid-market firms, this matters because the vendors you rely on for AP automation, close management, and ERP integration are increasingly being absorbed into larger players who are building proprietary AI stacks — stacks you'll have less visibility into and less leverage over. If you're currently evaluating or renewing contracts with outsourced finance tech providers, ask directly: who owns your AI model, where does your transaction data go during inference, and what happens to your data if they're acquired.
Before your next vendor renewal, get a written answer on data handling during AI inference — acquisition activity makes this urgent.AI Vendors Are Selling You Automation Built on Insecure Foundations — and You Have No Way to Know
The Cursor AI vulnerability is a symptom of a broader problem: the tools being used to build your accounting automation were never designed with financial data security as a constraint. Most AP automation and close workflow tools in the mid-market were built by small engineering teams moving fast, using whatever AI coding assistants were available — Cursor, Copilot, Codeium — and none of those tools have been seriously stress-tested for supply chain attacks targeting financial software specifically. The audit profession spent decades building controls around human error in financial processes; it has spent almost no time building equivalent controls around AI-assisted software development that underlies those same processes.
You've audited the output of your financial software for years. Nobody is auditing how that software was built.