Oracle: AI agents can reason, decide and act - liability question remains
- Reference: 1774460831
- News link: https://www.theregister.co.uk/2026/03/25/oracles_apps_and_agents_push/
- Source link:
Unveiled in London this week, Fusion Agentic Applications will be integrated with the Oracle Fusion Cloud Applications suite, covering financials, ERP, HR, payroll and supply chain management. Oracle argues it has a structural advantage here: the data needed to train and run these agents already lives inside its enterprise applications.
"Applications that can reason, decide, and act in pursuit of defined business objectives," is how Big Red's application development executive veep Steve Miranda framed the shift, a move away from process-focused software toward outcome-driven automation.
[1]
Oracle, for example, promises a Design-to-Source Workspace Agentic Application, which it says can work across engineering, supplier, and sourcing decisions to create one "coordinated and continuous process."
[2]
[3]
However, Balaji Abbabatulla, Gartner vice president and vendor lead for Oracle, was more measured, pointing to unanswered questions about how the technology will be implemented in an enterprise setting.
"Our position is that this sounds good, but be cautious. It doesn't necessarily look as glittery as it sounds. There are challenges under the hood which are not being overcome right now, but maybe over time," he said.
[4]
In January, Gartner said boards of global businesses are putting tech teams [5]under pressure to implement AI agents . Application, database, service layer, and cloud vendors are all scrabbling over the expected bonanza, trying to build influence over enterprise AI strategy.
Oracle’s pitch is to house AI agents within its enterprise application suite, and sell AI Agent Studio for Fusion Applications to help organizations build, connect, and run AI automation and agentic applications. Oracle has also launched an AI Data Platform to integrate data from different sources to build AI agents.
Gartner's Abbabatulla said that via the Platform Oracle wants to connect non-Oracle repositories, legacy applications - such as SharePoint repositories - and extract information from them. Although Big Red provides tools for data or technology experts to do that, it is not automated.
[6]
“There's no kind of autonomous way of synchronizing these different data repositories in the background,” he said.
[7]AI agents are 'gullible' and easy to turn into your minions
[8]Salesforce snaps up the team who built calendar app Clockwise to work on Agentforce
[9]SAP already shifting focus from ERP migration disaster in pursuit of AI-driven growth
[10]AI for software developers is in a 'dangerous state'
[11]UK Treasury not sure about ditching Oracle to join £1.7 billion shared services program it is funding
Building agents to run application-based processes will require a lot of work – and most likely spending money with Oracle to get the right engineering expertise, he added.
That's a hurdle for some large enterprises already invested in data platforms from Databricks, Snowflake, Cloudera or other vendors, with some initiatives harking back to the "big data" investment era. Abbabatulla sees Oracle's pitch as partly defensive, using data-in-context as an incentive to keep customers within its ecosystem.
“The transition overhead is massive, because these are investments people have made for years now,” Abbabatulla said. “This is unlikely to actually attract them to let go of this investment, but I'm sure there'll be organizations willing to try this in addition to some of those other investments they have made.”
Oracle and other vendors must still answer the question of who takes responsibility for AI decision-making should it go wrong, a problem The Register has been [12]raising for a couple of years .
If an AI agent makes a bad decision at scale and speed, cascading errors could spread before anyone notices. Oracle's answer so far is monitoring and audit tooling, but Abbabatulla is unconvinced: "I don't see a clear response from any vendor on the liability issue."
Mickey North Rizza, IDC group vice-president enterprise software, was more bullish, calling it a "significant shift" in agentic systems as they continuously complete work within the enterprise software system.
“Overall, this is a great move for Oracle positioning it as a market shaper towards the Agents as Apps. It won’t be the app with the best UI that does well, but rather the agent that reliably completes outcomes that are at scale, with trust and bring sustained economic leverage," she said.
With boards pressuring tech teams to deploy agents, Oracle, like every major platform vendor, is fighting for a piece of that pie. ®
Get our [13]Tech Resources
[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/applications&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2acRpFHW6kFV6-i3D_1kK0wAAAMc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/applications&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44acRpFHW6kFV6-i3D_1kK0wAAAMc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/applications&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33acRpFHW6kFV6-i3D_1kK0wAAAMc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/applications&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44acRpFHW6kFV6-i3D_1kK0wAAAMc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://www.theregister.com/2026/01/16/hyperscales_and_vendor_fund_trillion/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/applications&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33acRpFHW6kFV6-i3D_1kK0wAAAMc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2026/03/23/pwning_everyones_ai_agents/
[8] https://www.theregister.com/2026/03/20/salesforce_clockwise_team/
[9] https://www.theregister.com/2026/03/24/sap_commercial_focus/
[10] https://www.theregister.com/2026/03/18/ai_for_software_developers_qcon/
[11] https://www.theregister.com/2026/03/06/uk_treasury_matrix_nao/
[12] https://www.theregister.com/2023/09/28/llms_business_risks/
[13] https://whitepapers.theregister.com/
Re: Floating
> The models have no understanding of what they see, say, or do. They have no persistent world model, no capacity to verify their own outputs against reality, and no mechanism to recognise when they're confidently wrong.
Um, I've got some bad news about humans. Especially the ones who vote.
Re: Floating
100% this - there is no reasoning and there is no "meaning" to their outputs - I decide the meaning to the output of a band saw; its a useful tool at best. Arguably the use of sloppy anthropomorphic language - e.g. reasoning, attention, behaviour etc around these statistical text extruders makes it worse and is frankly very annoying, overloading concepts that have been used for centuries especially when dealing with real Artificial Intelligence research
Re: Floating
On the liability issue, it's more complex than user entered bad data into Excel and got a bad result. It's what the seller said the tool to do and what a 'reasonable' non technical person would understand that to mean. AI is tooted as being able to do fairly complex tasks accurately without (or with minimal) supervision by a person. Excel is not marketed with this expectation. Excel is marketed as a tool to do a variety of calculations and analysis after the data and formulas are entered (hopefully correctly) by the user. In the case of AI the legal issue is at best murky while in the case of Excel it is almost always going to be on the user for any errors.
Re: Floating
That's a fair distinction, but it actually makes the case worse for Oracle, not better. Think of it this way: if a company hires an employee, knows they need supervision, puts them in a decision-making role unsupervised anyway, and they make a costly mistake - the company is liable. Not the employee. Not the recruitment agency.
That's exactly what's happening here. Oracle is the recruitment agency. The AI agent is the employee. The enterprise knows - or should know - that the system is probabilistic, has no understanding of what it's doing, and will occasionally be confidently wrong. If they deploy it unsupervised in a consequential workflow anyway, that's on them.
The only thing the marketing language changes is whether Oracle shares some of that liability for overselling the candidate's abilities. But even in the best case for the buyer, that's a lawsuit after the damage is done - not a prevention mechanism. The enterprise still owns the outcome. Oracle sold them a tool and talked it up. What it does on the job is the employer's problem.
Re: Floating
Agents act on behalf of other entities but the legal liability for their actions stays with the entity on whose behalf they act. Its a pretty standard and well accepted definition.
For example, if the agent managing my rental property gets it wrong then I'm still the one legally liable to the tenant (this has happened). I can sue the agent for failing to meet their contractual obligations and maybe their liability insurance will kick in but that doesn't cancel out my liability to the tenant.
So yes absolutely the enterprise is on the hook. There should be no "it was a third party" excuses. Except that's precisely what they will trot out when it all goes wrong. Look a squirrel!
Larry Ellison
that's the Chinese agent who wants to form a surveillance state with his buddy's in the crime mob and palantir right?
Why are we listening again? Oracle has introduced some of the most hackable gear on the planet into business and industry for the purposes of spying by foreign agents.
Floating
AI agents can reason
Except AI agents can't reason. Generating an "internal monologue" is not reasoning - it's still most-likely-next-token prediction. The models have no understanding of what they see, say, or do. They have no persistent world model, no capacity to verify their own outputs against reality, and no mechanism to recognise when they're confidently wrong.
What Oracle is describing is automation with a probabilistic text engine at the decision point. That's not nothing - but calling it "reasoning" and "autonomous decision-making" is a marketing choice, not a technical description. A lookup table also "decides." A thermostat also "acts." Neither reasons.
The liability question isn't some minor open thread - it's already answered. If I use a circular saw and cut crooked, I don't sue Black & Decker. If my spreadsheet formula is wrong and I underpay staff, I don't blame Excel. The tool doesn't carry the liability. The person who selected, configured, and deployed the tool does. The reason nobody will say it out loud is that "we're selling you a tool you're fully liable for, that you can't fully predict or audit, and that will make decisions at a speed that guarantees damage is done before a human notices" is not a sentence that survives a procurement review. The liability question doesn't "remain." It's being deliberately left floating because the answered version kills the sale.