How OpenAI’s New Models Unlock Secure AI

If there’s one constant in AI, it’s that there is no quiet week.

Just last week, in my AI News of the Month edition, I told you we were expecting major releases from OpenAI in August. We’re only one week in, and we already have two — the long-awaited ChatGPT-5 and the release of OpenAI’s open-weight language models gpt-oss-120b and gpt-oss-20b.

The speed of change is relentless. In the space of a few days, what we thought we knew about how to work with AI can shift. New capabilities arrive, new constraints disappear, and in some cases, entire arguments against AI adoption dissolve almost overnight.

Both releases are significant, but for very different reasons. GPT-5 promises a smoother, more capable AI experience across reasoning, writing, and coding, with smart automation behind the scenes. The open-source release, on the other hand, answers a question that’s been holding finance back for years: how can we use AI for sensitive financial workflows without sending data outside our secure environment?

That’s why, even though GPT-5 grabbed the headlines, the open-weight models may be the bigger strategic story for CFOs.

The Balanced View: Why the Bigger Story for Finance Isn’t GPT-5

ChatGPT-5: Long-Awaited, Significant — But Incremental for Finance

The release of ChatGPT-5 was long-awaited, and OpenAI promised major improvements. GPT-5 is a unified model that decides in real time whether to give you a quick answer or switch into a deeper “thinking” mode for more complex tasks. This new auto-routing aims to make the experience smoother, reduce the need to manually switch models, and — likely — optimize costs and energy usage under the hood.

Where GPT-5 is Better

• Reasoning: GPT-5 thinking mode delivers more accurate, multi-step answers and is significantly less likely to “hallucinate” — factual error rates drop by up to 80% compared to earlier reasoning models.
  

• Coding: Better at complex debugging, front-end design, and larger repository work, with improved attention to layout and aesthetics.

• Writing: Handles more nuanced structures, such as free verse or structurally complex reports, while keeping tone consistent and aligned to instructions.

• Health & Knowledge Work: Scores higher on realistic, expert-defined scenarios in health and other economically valuable domains, acting more like a proactive partner that flags potential issues and asks clarifying questions.

• Safety & Honesty: Improved ability to recognize when a task is impossible or underspecified, and communicate limitations without misleading the user.

• Customization: Better at following user-set styles and instructions, with new personality settings for interaction style.

But that’s an OpenAI promise. In reality, after a few days of testing, the reviews across the community — and my own experience — are more measured. While GPT-5 is clearly more polished and capable, for most finance workflows it’s not a dramatic leap from GPT-4o.

It’s a meaningful upgrade for general-purpose work — but for CFOs, the truly strategic news this week comes from somewhere else.

Gpt-oss — A Turning Point for Finance

One of the biggest blockers to adoption in finance for a long time was IT telling finance teams AI wasn’t safe.

That was a fair point at the time, but now there’s a good answer to it.

OpenAI’s new gpt-oss family — a 20B and a 120B parameter model — changes the equation. These models deliver near-parity with some of OpenAI’s best proprietary reasoning models, can run locally or in a secure private cloud, and are available under the Apache 2.0 license.

Why does this matter for finance leaders? Because it directly addresses three of the biggest concerns we hear about AI adoption: data security, cost predictability, and long-term control.

1. Sensitive Financial Data Stays in Your Environment
Finance teams deal with information that simply can’t leave their systems — whether because of GDPR, SOC 2, industry-specific regulations, or internal policy. With gpt-oss, you can run advanced reasoning models entirely on your own infrastructure, meaning sensitive data never touches an external API.

2. CapEx Instead of OpEx
With proprietary models, your AI budget often grows month by month. Even a few cents per API call adds up when you’re processing thousands of reconciliations, compliance reports, or transaction analyses.

With open-source models, you make a one-time investment in hardware and deployment, and your ongoing usage cost approaches zero. This turns AI from an unpredictable operating expense into a fixed, depreciable asset — something CFOs can actually budget for and control.

3. Vendor Independence & Stability
We’ve all seen the risks of relying on a single AI vendor: price changes, feature removals, or model deprecations with little notice. With an open-weight model, the version you deploy is the version you keep — until you choose to change it.

4. Customization for Finance-Specific Workflows
Because gpt-oss models are fully customizable, you can fine-tune them with your chart of accounts, industry-specific risk frameworks, or proprietary forecasting logic.

5. Performance Without the Cloud Dependency
The smaller gpt-oss-20B can run on hardware with just 16GB of memory — enabling secure, on-premises deployment without high-end infrastructure.

When Open Source Makes the Most Sense

Open-source local models shine in predictable, repetitive processes that process large volumes of similar documents or information — especially when accuracy over time matters more than instant updates.

Best-Fit Finance Workflows:

• Document-heavy, rule-driven tasks (e.g., invoice classification, loan review)

• Internal knowledge Q&A (e.g., chatbot for finance procedures, historical audits)

• Compliance & policy monitoring (e.g., AML checks, policy enforcement)

• Historical data analysis (e.g., budget variance trends, M&A performance)

Where They’re Not Ideal

Running a model locally means you control it completely — but it also means it stays exactly as it is until you decide to update it. Unlike cloud models, which are silently improved in the background, a local model won’t get smarter on its own.

That’s fine for stable, internal workflows. But it’s not ideal for tasks where:

• The information changes daily, such as market news, FX rates, or commodity prices.

• You need to browse the internet to find or verify information.

• You depend on constantly refreshed external datasets like economic indicators, competitive intelligence, or new regulatory filings.

In these cases, a cloud-based model — or a hybrid setup that combines local AI for secure processing with cloud AI for live data — will deliver better results.

These new open-source models don’t just unblock use cases that were once shut down by IT — they also raise the bar for how CFOs should think about AI strategy. The real opportunity isn’t in choosing one model over another, but in designing smart, blended workflows that put the right type of AI in the right place. Use local, open-source models where privacy, predictability, and cost control matter most, and lean on cloud-based models for live data, complex integrations, or capabilities that benefit from constant updates. Done well, this hybrid approach gives finance leaders the best of both worlds: airtight security where it’s non-negotiable, and cutting-edge performance where it creates real competitive advantage.

Closing Thoughts

In less than a week, we went from anticipating “big AI updates in August” to having two releases that will shape how we use AI in finance.

ChatGPT-5 is a well-executed upgrade, improving reasoning, accuracy, and usability — and it will quietly save costs and energy in the background. But the open-source release is the one that could truly reshape AI strategy, giving finance teams private, predictable, and fully controlled AI capabilities.

Now is the moment to revisit your AI roadmap. Map out where open-source models can unlock secure, high-impact workflows, and where closed-source tools can keep you on the cutting edge. The finance teams who get this balance right will move faster — and more securely — than the competition.

If you have questions or want help designing these workflows, just reply to this email — I read and respond to every one.

See you next week with more actionable AI insights for finance leaders.

We Want Your Feedback!

This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!

Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation! 

Did you find this newsletter helpful? Forward it to a colleague who might benefit!

Until next Tuesday, keep balancing!

Anna Tiomina 
AI-Powered CFO