What changes for AI investing tools on 2 August 2026

The date

On 2 August 2026, the bulk of the EU AI Act becomes fully enforceable. The Act entered into force two years earlier, on 1 August 2024, with a staged timeline: prohibited practices took effect in February 2025, general-purpose AI obligations in August 2025, and the rest — including the rules that matter most for an AI investing tool — this summer.

If you are a European retail investor evaluating an AI research product, three provisions will materially change what that product is required to disclose, log, and label between now and August. They will also, indirectly, raise the floor of what reasonable users should expect from any AI investing tool — including products operated outside the EU. The Brussels effect tends to be quiet but durable. Tools that want EU customers will design to EU rules, and US-only AI products that decline to align tend to be the ones with regulatory exposure later.

What follows is a plain reading of the three provisions, what each one means in the context of a research product that uses large language models, and a checklist of seven things any user should be able to confirm about an AI investing tool in under a minute.

Article 4: AI literacy

Article 4 requires both providers (the companies that build AI systems) and deployers (anyone using those systems in their work) to ensure that staff have a “sufficient level of AI literacy.” It has been formally in force since 2 February 2025; national market surveillance authorities gain formal enforcement powers on 2 August 2026. The penalty tier sits at the intermediate level — fines up to €7.5 million or 1% of global annual turnover, whichever is higher.¹

“Sufficient” is proportional. The Act asks for literacy appropriate to the role: a compliance officer overseeing an AI-driven scoring model needs to understand the model's known failure modes and what its outputs do not tell them; a customer-support agent using an AI assistant needs to know when to flag a response as one not to act on. The standard is “to your best extent” — calibrated to the size, resources, and complexity of the deployment.

For an AI investing tool, two practical things follow. First, anyone the tool's company puts in front of customers — support, content, sales — needs working knowledge of how the underlying models can fail (hallucinated facts, stale data, drift, prompt-injection). Second, the tool itself should make the user's own literacy easier: showing what model produced what, naming its known limits, and explaining when an output should be treated as research rather than fact. Products that present model outputs as oracle and treat the user as a passive recipient now have a documented compliance gap.

Article 50: transparency

Article 50 is the provision most retail users will actually see. It requires two things that matter for an AI investing tool.

First, any system that interacts with people via text — chatbots, conversational research assistants, AI tutors — must clearly disclose that the user is interacting with AI, not a person.

Second, any system that generates synthetic text, image, audio, or video must mark its outputs in a machine-readable format and ensure they are detectable as artificially generated.² For an AI research tool, that means an AI-written paragraph in a research report cannot be presented as if a human analyst wrote it. The tool must surface, somewhere clearly visible to the user, that the output is AI-generated.

The European Commission published the first draft of a Code of Practice on marking and labelling AI-generated content in December 2025; a third and likely final version is expected by June 2026.³ The current draft proposes a standardised EU label — an “AI” visual mark (localised as “KI” in German, “IA” in French), with a taxonomy that distinguishes fully AI-generated content from AI-assisted content.

The practical test of compliance is uncomfortably simple. Open any analysis page in the tool, and ask whether you, looking at the page, can tell which sentence was written by a model and which was written by a person. If the answer is no, that tool has an Article 50 problem this August.

Annex III and the ESMA framing

The third change is more nuanced. Annex III of the AI Act lists the categories that count as “high-risk.” Three of these touch finance: credit scoring of natural persons, risk assessment and pricing in life and health insurance, and AI systems used to evaluate or classify the financial standing of individuals. Investment research and investment advice are not on this list.⁴

That distinction matters. A tool that scores stocks, analyses ETFs, or produces research reports is not, on its face, a high-risk AI system under the AI Act. It is not subject to the heavier obligations of Articles 9 to 15 — risk-management systems, training-data governance, immutable audit logs, human-override capability on consequential decisions.

But it is also not unregulated. The European Securities and Markets Authority published a Public Statement in May 2024 setting out its expectations of investment firms using AI when serving retail clients.⁵ On 26 February 2026, ESMA followed up with a supervisory briefing on algorithmic trading that explicitly addresses the interaction between algo-trading obligations and AI Act provisions.⁶ The combined position is clear. AI use in any investment service — research, analysis, portfolio management — has to satisfy the same MiFID II conduct-of-business obligations as a human would. Best interest of the client. Transparency. Recordkeeping. Suitability where investment advice is being given.

For an AI investing tool, the active compliance surface is the intersection of three regimes — AI Act transparency (Article 50), AI literacy (Article 4), and the MiFID II conduct obligations ESMA has confirmed apply to AI use in investment services. Plus GDPR, which has applied throughout. The August 2 date is when the AI-specific parts of that intersection become formally enforceable.

A buyer's checklist

The simplest way to evaluate an AI investing tool against the rules taking effect this summer is to ask, in order, the seven questions below. They are not exhaustive. They are the ones a serious user should already expect to be able to answer in a minute on the tool's website.

1. 1. Is every AI-generated output clearly marked as AI-generated? The Article 50 test. A research summary, a position analysis, a chat answer — each one labelled at the point of display, not buried in a Terms of Service link.
2. 2. Can you tell which model produced what? A tool that uses GPT-4o for one task and a fine-tuned variant for another should disclose both, and version them. “Powered by AI” is no longer specific enough to count as transparency.
3. 3. Does the tool publish a list of outputs it will refuse to generate? Article 4 literacy in practice means publishing the model's limits. A research tool that will not produce personalised investment guidance, ranked picks-of-the-week, or future price targets should say so — and explain why.
4. 4. Is there an audit trail of model outputs? If something the tool said today contradicts something it said last week, you should be able to see both, with timestamps. Tools that overwrite history are tools that are very difficult to evaluate.
5. 5. Does the company publish information about model risks and limitations? A model card, a methodology page, or a disclosure document — under any name — that names the known failure modes and the data the model depends on. If the only document you can find is the marketing landing page, the literacy gap is the tool's, not yours.
6. 6. Does the tool stop short of personalised investment advice? This is a MiFID II question, not an AI Act question, but the two are entangled in practice. A tool that issues research is generally not regulated. A tool that takes your profile and names a specific instrument as the one for you generally is — and unless the company holds the required licence (BaFin, FCA, or SEC equivalent), it has a problem. Personalised model portfolios of named tickers, and any phrasing that pairs a specific instrument with the user's personal investment profile, are the trigger lines.
7. 7. Is the legal basis for processing your data clear? GDPR overlaps the AI Act here. A tool that uses your portfolio data to fine-tune its models should tell you, ask for separate consent where relevant, and explain how to opt out.

These are questions a serious tool should already be designed to answer affirmatively. None of them require a law degree to evaluate. The pattern to watch for is the answer “yes, but it's complicated” — which is usually a sign the tool is preparing answers it did not expect to have to give.

What doesn't change

A few things are worth being clear about, because the headlines around the August 2 date have been confused.

The AI Act does not bar AI tools from producing investment research. Research is a permitted activity. Opinion writing is a permitted activity. Analysis is a permitted activity. The Act regulates how AI is built and presented, not which subjects can be written about.

The AI Act does not require an AI investing tool to obtain a regulatory licence simply because it uses AI. A research product is still a research product. A product that crosses into personalised advice is a regulated activity, but that has been true under MiFID II since 2018 — the AI Act does not change the categorisation.

The Act does not pick a winning model architecture, or bar use of any particular foundation model. Products using GPT-4, Claude, Gemini, Llama, or anything else are equally subject to the same transparency, literacy, and (where applicable) high-risk obligations. The model provider has obligations too, but most of those came into force separately, on 2 August 2025, under the General Purpose AI rules.

What the Act does do, on 2 August 2026, is shift the cost of opacity. Tools that have been quiet about their model use, their failure modes, and the line between AI-generated and human-authored content will start paying for that quiet — first in trust, then in enforcement.

The trust gap

The Investing.com 2026 retail survey put a useful number on the asymmetry: roughly 62% of retail investors now use AI tools in their research process, but only 23% trust the outputs without verifying them against other sources.⁷ The remaining 39% — the use-but-verify group — describe behaviour the AI Act is, in effect, writing into law. Read the model's output, treat it as one input among several, check the source.

The provisions taking effect on 2 August turn that intuition into a documented standard. They will not make AI investing tools materially harder to use. They will make it materially harder to operate one that hides what it is doing. The trust gap closes from the supply side, not the demand side.

For users, the meaningful change is straightforward. Before this summer, asking an AI investing tool to label its outputs, name its models, and disclose its limits was a reasonable request. After this summer, in the EU, it is the default. Tools that meet the bar quietly will not advertise it. Tools that do not will either fix the gap or stop serving EU customers.

Acutic is being built around this bar. The AI transparency page names every model the system uses and what each one is used for. Every analyst output carries an AI-generated label at the point of display. The methodology is public. The product produces research and analysis, not personalised investment advice — and the user makes every final decision.

Notes

1. ¹AI Act Article 4, with penalty tiers under Article 99. Full text at artificialintelligenceact.eu/article/4.
2. ²AI Act Article 50(2). Machine-readable marking is the more technical of the two requirements; “detectable as artificially generated” is the user-facing one.
3. ³European Commission, Code of Practice on marking and labelling of AI-generated content, first draft published 17 December 2025.
4. ⁴AI Act Annex III. The financial-services items sit in section 5 of the Annex and cover creditworthiness, insurance pricing, and evaluation of financial standing.
5. ⁵ESMA, Public Statement on the use of AI in investment services, ESMA35-335435667-5924, May 2024.
6. ⁶ESMA, Supervisory briefing on algorithmic trading, 26 February 2026.
7. ⁷Investing.com, How Retail Investors Are Using AI in 2026, retail-investor survey, April 2026.

Further reading on how Acutic approaches the same questions in product: see the public methodology page, the AI transparency page, or the previous essay, portfolio rules you actually keep. Request early access.

Acutic provides investment research and educational content. It is not investment advice. Acutic operates as a non-personalised investment research and analysis service under MAR Art. 20 / § 85 WpHG. Past performance does not predict future results.