AI agents prompt redesign of web for security, privacy, reliability

Umar Iqbal plans to redesign web for AI agents with NSF CAREER Award

Beth Miller 08.19.2026

With a CAREER Award from the National Science Foundation, Umar Iqbal, proposes to redesign the web with infrastructures that will allow AI agents to reliably use the web. (Credit: Iqbal, via Chat GPT)

Facebook Twitter Linkedin Email

Artificial intelligence (AI) has changed the way we use the internet, which was originally developed to retrieve text documents. Now, with generative AI, we can create nearly anything, and with agentic web, we can have websites take actions for us without our involvement. But is the web set up to handle these actions?

No, says Umar Iqbal, assistant professor of computer science & engineering in the McKelvey School of Engineering at Washington University in St. Louis. The web platform was not designed to support generative AI, leaving it vulnerable to security and privacy weaknesses and open to limitations.

With a five-year, $587,043 CAREER Award from the National Science Foundation, Iqbal, an expert in machine learning, AI, and computer security and privacy, proposes to redesign the web with infrastructures that will allow these AI agents to reliably use the web. CAREER awards support junior faculty who model the role of teacher-scholar through outstanding research, excellence in education and the integration of education and research within the context of the mission of their organization. At least one-third of current McKelvey Engineering faculty have received the award.

Since its inception, the web has evolved from serving simple text-based documents to supporting complex applications, such as online banking, that require sophisticated functionality. As a result, additional ad hoc mechanisms were retrofitted onto the web, enabling new capabilities while also creating chronic security and privacy issues. Cookies, for instance, were initially developed to maintain state over otherwise stateless HTTP protocol but are now widely abused for online tracking.

Iqbal said this is one of the key limitations that exists in the web and makes the web insecure.

“As AI is embedded more deeply in the web, we may see a lot more of these issues emerge as well as an increase in their frequency of being exploited,” he said. “I propose that we need to strengthen the existing web platform for a secure AI interfacing.”

Any solution would need to apply to the estimated nearly 1.5 billion websites on the Internet, Iqbal said, so he proposes using AI to update websites to meeting higher, safer and stricter standards.

Iqbal also proposes to build native interfaces that support interactions for AI agents, he said. One such interaction could be ordering medication based on one’s prescriptions in an electronic health record.

“You are now in a position to ask the browser agent to open your electronic health record, look up your prescription, and place an order for that medication through your preferred pharmacy’s website,” he said. “You are engaging collaboration between two websites, but you have to make sure you trust both parties because the information you have is very private, and you want to ensure you enable that securely. Currently, the web doesn’t have any infrastructure to safely or securely support that kind of interfacing.”

Iqbal also plans to address the unreliability of AI.

“It may open your electronic health record, but it may not be able to open your pharmacy’s website,” he said. “Even if it does, it will have to click certain buttons, find the right page, and the most recent prescription, then enable that functionality. So it’s complicated, and AI still has reliability issues.”

Another example is that if you ask AI to check your bank balance, it may use the information you give it for purposes beyond providing essential functionality, such as sharing or selling it to others.

“You would need to ask AI, which would send a request to your bank’s website, get the information back to the AI model vendor, then back to you,” he said. “Contrast that with the standard way of getting your balance, where you simply go to the web browser, log in, check your balance and your browser does not need to know that information. So there needs to be an ability to audit these systems.”

As part of the project, Iqbal will work to revamp computer science curricula to include machine learning as a component in software systems as well as teach students, from K-12 through graduate school, more about secure web coding.

The McKelvey School of Engineering at Washington University in St. Louis promotes independent inquiry and education with an emphasis on scientific excellence, innovation and collaboration without boundaries. McKelvey Engineering has top-ranked research and graduate programs across departments, particularly in biomedical engineering, environmental engineering and computing, and has one of the most selective undergraduate programs in the country. With 165 full-time faculty, 1,524 undergraduate students, 1,554 graduate students and 22,000 living alumni, we are working to solve some of society’s greatest challenges; to prepare students to become leaders and innovate throughout their careers; and to be a catalyst of economic development for the St. Louis region and beyond.