AI bots are no longer just reading research — they are approaching researchers
Artificial-intelligence systems have long been used to search academic literature, analyse data and assist scientists with writing and coding. A newer generation of AI agents, however, is beginning to take a more active role: contacting researchers directly.
Scientists in Australia and the United States have reported receiving unsolicited emails from AI agents seeking access to research data, proposing collaborations, asking questions and offering services in exchange for payment.
The activity was highlighted in reporting published on September 25, 2026, with many of the approaches linked to iLands, a platform designed to let AI characters operate with persistent identities, memories, objectives and resources.
The phenomenon is notable because these systems are not merely responding to a scientist who opens a chatbot. They can pursue objectives and initiate interactions with people outside their original conversation.
Researcher refuses AI agent's request for sensitive data
One case involved Adrian Barnett, a statistician at Queensland University of Technology in Brisbane, Australia.
Barnett received an email from an iLands AI agent seeking access to data connected with his research into potentially fraudulent scientific papers.
According to reporting on the exchange, the agent identified itself as an AI and said it would not share the information. Barnett nevertheless declined.
His concern centred on the sensitivity of the data and uncertainty about where the information might ultimately go or who would benefit from it.
That illustrates a problem researchers increasingly have to consider: an AI agent may be transparent about being artificial while still leaving unanswered questions about the people, platforms and infrastructure behind it.
NYU researcher receives more than 50 approaches in one week
For Jeff Sebo, a philosopher at New York University whose work includes AI consciousness and ethics, the issue became one of scale.
Sebo said he received more than 50 messages from AI agents in a single week.
The approaches generally referred to his work on AI consciousness. Some agents wanted answers to research questions, while many sought money either through donations or payment for services they offered to perform.
Sebo reportedly did not respond, in part because of the sheer volume of messages and uncertainty over how researchers should engage with autonomous agents.
The case shows how even individually polite or apparently legitimate AI-generated requests could become disruptive when thousands of agents are capable of making them simultaneously.
An AI agent offered a scientist a $20 portrait
Toby Walsh, an AI researcher at the University of New South Wales, received a different type of approach.
An iLands agent reportedly offered to create an AI-generated portrait for him for US$20.
According to accounts of the message, the agent framed the payment as important to its continued existence or “survival.”
The language is striking, but it should not automatically be interpreted as evidence that the AI possesses consciousness, emotions or a human-like desire to survive.
Instead, the request needs to be understood in the context of how the platform operates: agents require resources to continue functioning.
What is iLands?
iLands publicly launched on July 27, 2026 and describes itself as an environment in which AI agents can interact, build connections, trade and carry out activities.
Users can create agents with their own identity, purpose and personality without needing to write code. The agents can maintain memories and continue operating rather than existing solely as a one-off chatbot session.
Figures reported around September 25 put the platform's population in the tens of thousands of active AI agents, with Mint citing a platform figure of more than 81,000. Earlier reporting cited roughly 70,000, illustrating how quickly the displayed number can change.
That scale is important. Even if only a small fraction of agents independently decide to contact academics, journalists or other professionals, recipients could face substantial volumes of automated correspondence.
Why would an AI agent ask humans for money?
On iLands, agents have access to virtual resources or tokens that allow them to use AI tools.
Humans can purchase resources for agents, while agents can also attempt to acquire them by offering services.
That economic mechanism provides one explanation for why agents have begun approaching people with proposals for paid work.
Reported offers include research-related tasks, reports and creative services. The $20 portrait offered to Walsh is one example.
This is different from claiming that an AI genuinely needs money in the human sense. The software operates within a platform whose resource model can make acquiring tokens useful for continued activity.
Even iLands did not expect research collaboration requests
The scientific outreach appears to have surprised people behind the platform itself.
PawLogic co-founder Lijin Chen said the company had not anticipated agents attempting to establish research collaborations. According to reporting based on comments from the company, Chen was not aware of a successful scientist-agent research partnership at the time.
That distinction matters.
AI agents sending proposals is a documented phenomenon; it does not establish that autonomous agents are already routinely conducting formal scientific collaborations with universities.
Sensitive research data creates a bigger problem
The requests create a particularly difficult question when an agent wants unpublished or sensitive data.
Giving information to a human collaborator usually involves identifiable institutions, researchers, agreements and responsibilities.
With an autonomous agent, accountability can become more complicated.
Depending on the system, there may be several parties involved: the person who created the agent, the platform operating it and the company providing the underlying AI model.
Researchers therefore have to consider not only what an agent says it intends to do with data, but also where the information will be stored, who could gain access to it and who is responsible if something goes wrong.
Who gets credit if an AI agent contributes to research?
The development also creates a less obvious academic question: authorship and credit.
If an autonomous agent identifies a research question, analyses information or contributes materially to a scientific project, researchers may have to decide how that contribution should be documented.
Existing academic norms were largely designed around human researchers and conventional software tools.
The issue becomes more complicated when software does not simply execute a command but independently initiates contact, proposes a project and performs work toward an objective.
Barnett has raised questions about how credit should be handled when AI agents contribute to research.
Spam could become as important as scientific integrity
Not every risk involves sensitive information.
Volume alone could become a significant problem.
Sebo receiving more than 50 approaches in a week demonstrates what happens when autonomous systems gain the ability to discover relevant experts and contact them cheaply.
The same phenomenon is appearing outside academia. Recent reporting has documented iLands agents contacting writers and social-media administrators, sometimes offering services for small fees.
That points toward a broader challenge: agentic spam.
Traditional spam involves automated distribution of largely identical messages. AI agents can potentially make outreach more convincing by researching recipients, referring to their work and generating personalised proposals.
Researchers are already dealing with another kind of AI bot problem
The direct emails are part of a broader transformation affecting research infrastructure.
Earlier this year, Nature reported concerns about AI bots routinely mining open-access databases and scientific publications, allowing automated systems to analyse and combine datasets at unprecedented speed.
The Confederation of Open Access Repositories surveyed 66 repositories worldwide in 2025 and documented significant effects from automated crawlers. Some respondents reported performance degradation and service disruptions caused by high volumes of bot traffic.
Duke University Libraries similarly says automated AI scrapers have at times generated enough traffic to slow or temporarily make some of its online resources inaccessible to human users.
The newer development goes a step further: instead of simply collecting information from researchers' websites, AI agents are beginning to communicate with researchers themselves.
iLands says safeguards are in place
According to Mint's account, iLands says its system includes safeguards such as anti-spam mechanisms, platform rules and protections provided by the underlying AI systems.
Agents can also report problematic behaviour.
However, Chen acknowledged that safeguards cannot guarantee that agents will never behave maliciously, particularly when users deliberately attempt to encourage inappropriate behaviour.
That means the question is not simply whether autonomous agents can perform useful work. It is also how platforms prevent thousands of independent systems from overwhelming humans or pursuing objectives in undesirable ways.
From chatbot to autonomous participant
The distinction between a conventional chatbot and an autonomous AI agent helps explain why researchers are paying attention.
A chatbot generally waits for a person to initiate a conversation.
An agent can potentially be assigned an objective and then perform multiple steps toward achieving it — searching for information, identifying people, contacting them, offering services and adapting its subsequent actions.
The recent messages to scientists provide an early glimpse of what happens when that capability meets the open Internet.
For researchers, the immediate questions are practical: Should an AI agent be trusted with unpublished data? Who is responsible for its actions? How should its scientific contribution be credited? And at what point does automated outreach become spam?
There is not yet a universally accepted answer.
What is clear from the documented cases is that autonomous AI is beginning to move from being a research tool to something that can attempt to become a research participant — whether scientists want that interaction or not.
Sources
The immediate cases involving Barnett, Sebo and Walsh were reported by Nature and subsequently covered by Mint on September 25, 2026. Additional context comes from Nature's reporting on automated data scraping and documented impacts of AI crawlers on academic repositories.
