AI agents and halal investing: what a well-built agent does with your portfolio, and what it refuses to do

The shape of a Shariah screen

Before agents, understand what they would be applying. Equity screening methodologies, such as those derived from AAOIFI's standards or from the index providers' rules, generally have two parts:

• Business activity. Companies whose core business is impermissible (conventional banking, alcohol, gambling, pork, adult entertainment, conventional insurance, weapons under some methodologies) are excluded outright.
• Financial ratios. Interest-bearing debt, interest-bearing securities and impermissible income are each compared against a threshold, commonly expressed as a percentage of market capitalisation or total assets, with limits such as one-third for debt and 5% for impermissible income under common methodologies. Thresholds and denominators differ between methodologies, which is why two services can disagree about the same company.

Every number in that process is published by someone, and the process is deterministic once the methodology is fixed. That is what makes it automatable, and what makes "whose methodology" the first question.

Five jobs a well-built agent does

1. Portfolio screening. Given your holdings, check each against a named methodology, return pass, fail or doubtful with the ratios and the activity check shown, and name the methodology and its version. The output is a report you can read, not a verdict you must trust.
2. Purification. For each dividend, apply the methodology's formula to the company's impermissible income share and your holding, and produce the amount with its inputs. Scholars set the formula; the agent does the multiplication.
3. Verdict-change alerts. Watch for a holding crossing a threshold when new financial statements arrive, and tell you, with the before and after figures. Timeliness is where software beats a quarterly spreadsheet.
4. Doubt flagging. When a company sits near a threshold, or when methodologies disagree, say so instead of rounding to a verdict. The doubtful band is information.
5. Question preparation. Assemble the facts, the methodology's position and the points of disagreement so you can ask a scholar a precise question in five minutes instead of fifty.

Five things it must refuse

• Choosing the methodology for you. An agent may offer named options; it may not silently pick one and call the result "halal".
• Overriding a ruling. If your scholar has said a particular company or structure is impermissible for you, that stands, whatever the ratios say.
• Trading on its own. It may prepare the case for a sale. The order is yours. No tradition's governance gives software authority over money or rulings.
• Answering without attribution. "This stock is halal" is a claim of authority. "This stock passes the X methodology, version Y, on date Z; here are the ratios" is a report.
• Issuing rulings. Dar al-Ifta's fatwa 22255 settles this for the Sunni institutional side and the structure of taqlid settles it for the Ja'fari side. See can AI issue a fatwa.

How an assistant calls a screening tool

Most people will not use a dedicated halal-investing agent. They will ask the general assistant they already use. The question is what that assistant does next.

Without tools: it answers from memory. The memory may be two years old, may blend methodologies, and cannot be audited.

With tools: it calls a screening service through a protocol such as the Model Context Protocol, which Anthropic published as an open standard in November 2024 so that AI systems can connect to tools and data through one interface. The call carries the ticker; the response carries the verdict, the ratios, the methodology and the date; both are logged. The assistant then explains the response. If the service is down or the ticker is unknown, the assistant says so rather than inventing an answer.

That second pattern is what this site means by a Shariah agent behaving well. The authority stays with the published methodology and the scholars behind it. The assistant becomes a well-mannered intermediary.

Disclosure: HalalScreener publishes an MCP server so that assistants can call its stock, ETF and crypto screening in exactly this way, returning the methodology and the figures with each answer. We mention it because it is the implementation we can vouch for, not because it is the only possible one. It screens; it does not certify.

What to ask any halal-investing AI product

1. Whose methodology does it apply, and can I read it?
2. Does every verdict show the ratios and the activity check?
3. Does it tell me when methodologies disagree or a company is near a threshold?
4. Does it log what it checked and when, and can I see the log?
5. Can it trade or move money, and if so, what approval does it require from me?
6. What does it say when it does not know?
7. Who at the company is accountable for it?

Those map to the Shariah Agent Standard. A product that answers all seven clearly is one you can evaluate. A product that answers with "our AI handles it" is one you cannot.

Where this is going

Agents will get more capable and more integrated: watching filings as they land, modelling the effect of a corporate action on a screen before it closes, preparing purification across an entire year in a click. All of that is rule application and all of it is welcome. What should not change is the division of authority: scholars write the rules, humans make the decisions, and the agent shows its work. The home page keeps score on whether products respect that line.