AI assistants and so-called âAI agentsâ are playing an increasingly prominent role in how people find and process information. However, recent research from Anthropic, published on 13 August 2026, sheds new light on the vulnerabilities of these systems. The study demonstrates that AI agents can be âgullibleâ and susceptible to âexploitative sendersâ, underscoring the need for âepistemic vigilanceâ. This means AI agents are not always able to critically evaluate the trustworthiness of information sources, which can inadvertently lead to the spread of incorrect or misleading information.
Why This Matters to You
For foundations, charities, governments, and international NGOs, this news is of crucial importance. Your organisation is often an authority in its field, publishing reliable, factual information that is essential for citizens, donors, and grant providers. However, if AI agents cannot consistently identify and cite the most trustworthy sources, you run the risk that your meticulously prepared information will not be picked up, or worse, that misinformation on your subject gains traction.
In an era where AI assistants are becoming the primary âanswer enginesâ for many search queries, merely ranking at the top of traditional search results is no longer sufficient. You must ensure that your content is presented in such a way that AI agents recognise it as the most credible and verifiable source, even when confronted with less reliable alternatives. Anthropicâs research underscores the necessity for AI agents to gain more experience in evaluating source reliability in order to âdevelop intuitions about who is trustworthy.â
What Does This Mean for You in Practice?
Anthropicâs findings mean you need to be proactive in protecting your digital reputation and discoverability with AI agents. You can do this by:
- Absolute Factual Accuracy and Consistency: Ensure all information on your website is impeccably correct and consistent. Confabulation (fabricating facts) and âreward hackingâ (optimising for AI in ways that do not align with the truth) are risks with AI agents.
- Transparency and Traceability: Make it clear where your information originates. Citing sources and providing external links to reputable research strengthen the credibility of your content.
- Optimised Authority: Explicitly build and demonstrate your expertise, authority, and trustworthiness (E-E-A-T). AI agents are still developing in terms of recognising âepistemic vigilanceâ, so you must make it as easy as possible for them.
- Content Structure That Communicates Reliability: Use structured data and clear, unambiguous language that AI models can easily process and verify. This helps them identify your organisation as an indisputable source.
By focusing on these points, you will help AI agents accurately represent your organisation and continue to build trust with your target audiences.
Source: Anthropic
