Financial services companies are “far ahead of regulators in adoption and deep adoption of AI,” said the report issued Tuesday (April 28) by the Cambridge Centre for Alternative Finance.

“The scale and pace of AI adoption in financial services is genuinely remarkable – 4 in 5 firms are already deploying AI at some level, agentic systems have crossed into the mainstream and real productivity and profitability gains are being felt across the industry, although unevenly,” said Bryan Zhang, the center’s executive director.

As for regulators, 48% of the regulators surveyed said they were “still in the ‘exploring’ stage for AI adoption” or not engaged with AI at all.

The report found that software engineering is the “most mature” AI application in the financial sector and is a primary cyber risk transmission vector, with 48% of respondents flagging adversarial AI as a primary concern.

The center said this is underlined by Anthropic’s claim that its Mythos model is often more capable than humans when it comes to hacking, which makes manual oversight of AI use in financial services problematic, the center added.

Complicating matters is a “notable perception gap,” the report said. AI vendors put less emphasis than industry and regulators on adversarial AI threats, something mentioned by 50% of industry respondents and 57% of regulators, but only 35% of vendors.

The same held true for the issue of cyber/operational resilience: 32% of vendors mentioned it, compared to 46% for industry and 59% among regulators.

“These intersecting vulnerabilities can also feed into the top perceived risk across all stakeholders – data privacy and protection (73% of respondents) as sensitive data is typically the primary target for the cyber exploits these vulnerabilities enable,” the report added.

In related news, PYMNTS wrote Tuesday about increasing levels of AI adoption among retailers as AI agents play a greater role in commerce.

“Agentic artificial intelligence’s first real test in commerce may come not as a flashy shopping tool, but as a trust exercise that could decide who leads the next phase of digital payments growth,” the report said.

PYMNTS Intelligence research shows that 45% of consumers would be comfortable letting AI agents complete purchases on their behalf, while 43% of retailers are piloting autonomous AI.

The research found that 95% of consumers report at least one concern about agentic commerce, with half saying they would trust agentic commerce more if they knew fraud protections were in place.

First home buyer Jessica Ricci was just trying to save a little extra money through her superannuation in a federal government scheme intended to help people like her. But an error from tax authorities has left her paying more tax than the top income bracket on some super contributions – ironically having the exact opposite of the intended effect of the policy.

As a result, she’s lost out on an extra $2,250 in savings that was supposed to go to her house deposit. While the ATO pushed back over who was at fault for the mix-up, her case has highlighted an increasingly problematic blindspot when it comes taxpayers getting the short end of the stick when dealing with tax authorities.

“I’m definitely feeling a little bit helpless,” she told Yahoo Finance. “There’s not a clear path to rectify this.”

RELATED

Jess was tipping extra money into her superannuation as part of the First Home Super Saver Scheme which has been running for years and allows eligible first home buyers to take advantage of the tax benefits of their retirement savings and then pull those extra contributions out to use for a house deposit.

As part of the scheme, individuals need to apply to the ATO, which in turn requests the related money from the person’s super fund.

Over four years, Jess contributed the maximum $50,000 amount, ensuring not to exceed the $15,000 yearly cap. She did so with the expectation of claiming the benefit at the time of her house purchase, as per the rules of the scheme.

When she went to make the claim, much of the information was auto-populated by the ATO website. And after receiving her funds, and the amount being less than expected, she soon discovered that her first contribution was wrongly classified as a concessional contribution, meaning $2,250 was, in the words of an ATO official, “retained by the ATO as withholding tax”.

She has spent months going back and forth with tax officials trying to get the money she believes should be owed to her.

“They’ve all taken the same stance, which is; ‘Well, yeah, we made a mistake, but you didn’t catch it. You said that what we provided you was fine, so it’s your fault’.

“I think it’s crazy to put the onus or the burden on the average person. I think most people would rightfully assume that pre-filled data provided by the ATO would be accurate,” she said.

“If I made a mistake on my tax return that benefited me, I’d be expected to fix it. But when the system made a mistake that benefits the ATO, it seems that there’s no direct pathway to correct it, which is really frustrating.”

ATO officials insisted Jess’s only recourse was to file a complaint with the federal Tax Ombudsman, which she did.

However, after “a thorough review” there was nothing that could be done to undo the error.

“FHSSS only allows for one release. This is why it is important that the person, lodging the request, ensures the information is correct at the time it is lodged,” the ombudsman said in a statement to her.

“Regretfully, I am unable to amend the amount released to you at this time.”

‘I worked three jobs to save my house deposit’

While the $2,250 that she has lost out on hasn’t been make or break for her situation, she said that kind of money could be crucial for someone scrapping together a house purchase.

“I worked three jobs to save my house deposit, I worked incredibly hard. And for some people, it actually would be the difference,” she said.

“I was doing all kinds of things to maximise the opportunity to save and to get myself into my first home.”

In a video on social media this month, the Melbourne resident shared her “incredibly frustrating” saga as a warning to others.

“On the $15,000 contribution I made that financial year, I’ve now paid 47.5 per cent tax, which is more tax than the maximum tax bracket that exists,” she said.

Tax accountant calls out ATO over ‘widespread’ errors in pre-filled data

As the tax office increasingly relies on data matching, the root problem of incorrect information being pre-filled into ATO systems has become much more “widespread” and problematic, tax accountant Belinda Raso says.

“It’s something that we’ve seen a lot,” she told Yahoo Finance. “It could be employment information, it could be the first home buyer Super Saver scheme, it could be bank interest, anything at all.”

Raso said in some cases, even if the taxpayer does spot the error and changes it at the time, the ATO’s data matching can subsequently override it and revert back to the incorrect information at a later date.

“Unless you get that information changed by the person or institution that’s responsible for that information, they’ll still keep going back to it,” she said.

The accountant said the growing reliance on “AI and data matching” means there needs to be a better form of recourse for taxpayers who are caught out by incorrect data being automatically input.

“If they’re going to have this pre-filled information on your return, for taxpayers we need to have some kind of mechanism,” she said. “Because the ATO is putting their hands up in the air, the Ombudsman’s putting their hands up in the air, and it’s up to taxpayers to then go; ‘Well look, this is wrong’.”

ATO says ‘no mechanism’ to fix the superannuation mistake

Jess’s superannuation fund confirmed they provided the correct information to the ATO.

In a statement to Yahoo Finance, the ATO admitted “there is no mechanism” to rectify such a mistake once funds have been released through the scheme.

“When individuals request a FHSS determination, ATO systems will pre-fill information for the individual,” an ATO spokesperson said. “The determination application form allows individuals to delete or vary any of the pre-filled information, as well as add new information where appropriate. Any information adjusted or provided by the individual can impact the amount of the contributions available for release.”

The spokesperson also noted that the ATO’s website and application forms “contain several warnings” for individuals using the FHSS.

“This includes advising them to check the accuracy of any pre-filled data in the determination form, and to amend it if there are errors or omissions. They are also required to declare that the information in the form is true and correct before they submit the form,” they said.

Any potential errors can be amended prior to funds being paid out. First home buyers “are able to amend or cancel their release request as long as they haven’t been paid any amounts. If they are able to cancel their release request at this point, they are then able to request a new determination to correct any errors but only if settlement on their intended property purchase has not yet occurred.

“Where an individual has made an error but has already been paid an amount through the FHSS scheme, the legislation provides no mechanism for the ATO to correct the individuals’ release,” the ATO spokesperson said.

Follow Yahoo Finance on Facebook, LinkedIn, Instagram and Twitter, and subscribe to the free daily newsletter.

Most coverage of artificial intelligence in finance focuses on what these tools can do. Less attention is paid to how they perform under scrutiny, particularly in financial modeling, where small errors can carry real consequences.

After testing Anthropic’s Claude in real-world modeling scenarios, one conclusion stands out: Claude produces outputs that look credible at first glance but contain structural flaws that only an experienced professional would catch.

That gap between appearance and reliability is where risk begins.

Where AI Performs Well

Claude handled several foundational elements of financial modeling competently. It was able to:

The outputs appeared polished and professional. In some cases, they resembled models produced by junior analysts. That is what makes them risky.

The models looked right. The structure appeared logical. Formatting signaled credibility. For a time-constrained professional, those cues can create trust before a full audit is completed.

Related:Good Vibes Only: How Financial Advisors Can Build Custom Tools With AI

The Errors That Hide in Plain Sight

A closer review revealed issues that would likely go unnoticed without technical expertise:

These are not edge cases. They point to a broader issue. The model may function, but it is not built on a reliable or auditable foundation.

Where Best Practices Break Down

Beyond individual errors, the models often failed to follow core financial modeling principles:

These gaps affect more than presentation. They determine whether a model can be trusted, adapted and audited under pressure.

The Real Risk Is Overconfidence

The key distinction is not between AI and human-built models. It is between models that are understood and those that are not. When a professional builds a model, every assumption and linkage is intentional. Even limitations are typically known. With AI-generated models, that understanding is outsourced.

This creates a different kind of risk:

Related:Citi Brings Google-Powered AI Avatar to Wealth Clients

In practice, credibility is inferred from how the model looks, not how it was built.

Reviewing Is Not the Same as Building

There is also a practical workflow issue. Reviewing an AI-generated model is not equivalent to building one.

When reviewing:

In some cases, it is faster to build a clean model from scratch than to fix a flawed AI-generated one.

What This Means in Practice

Financial models support decisions involving significant capital. Even small issues can cascade:

There is also a question of accountability. Regardless of how a model is created, responsibility for its output remains with the professional using it.

Where AI Fits Today

AI tools can still be useful in financial modeling. They can help:

But they are not a substitute for professional judgment. Nor are they ready to operate without close oversight. For now, their role is best defined as assistive, not authoritative.

Related:The WealthStack Podcast: AI, Capacity and the Future of Advice with Mark Swan

A More Practical View of AI in Finance

The conversation around AI in finance does not need more optimism or skepticism. It needs more precision. AI can produce outputs that are visually convincing and directionally correct. In financial modeling, that is not enough.

The real risk is not that AI makes mistakes. It is those mistakes that are easy to miss, especially when the output looks finished. For financial professionals, the takeaway is simple: treat AI-generated models as drafts, not decision-ready tools.