It’s not just job loss and data centers that threaten our future well-being. Tech entrepreneurs are eager to embed AI into as many businesses and industries as possible. Here are some things we might look forward to as our key endeavors embrace AI.
Agriculture
AI systems are aimed at everything down to how we grow our food. They can analyze drone and satellite imagery to measure soil moisture and nutrient levels plant-by-plant. This enables targeted, micro-dosing of fertilizers and water, dramatically reducing chemical runoff. Autonomous tractors and smart implements use computer vision to distinguish between crops and weeds, applying herbicides only where needed. Robotic harvesters can even identify the ripeness of produce before picking. Machine learning models process leaf and plant images to identify early signs of pests, mold, or disease, allowing farmers to intervene before an entire harvest is compromised. AI aggregates weather forecasts, historical yield data, and current market demands to help farmers forecast commodity prices and determine the absolute best times to plant and harvest. Even wearable devices and barn sensors collect physiological data from animals. AI pattern recognition can detect illnesses, lameness, or changes in behavior early, allowing for timely veterinary intervention.
Human participation in agriculture is the foundational driving force that transforms natural environments into sustainable food systems. People provide the essential decision-making, labor, innovation, and adaptive management needed to navigate changing climates, ensure food security, and balance economic growth with environmental conservation. The risks of AI in agriculture include cyber threats to autonomous machinery, environmental degradation from prioritizing short-term yields, and increased socioeconomic gaps. Heavy reliance on these tools can also erode traditional farming instincts.
Education
In education, AI is envisioned as an aide to teachers by automating routine, time-consuming tasks like grading, generating practice questions, and drafting lesson plans. This reclaims time for 1-on-1 mentoring and relationship-building. Intelligent systems analyze student performance in real-time, helping educators quickly spot students who are struggling or identify concepts that need to be re-taught. For students, programs like Khan Academy Khanmigo act as 24/7 AI tutors, providing students with immediate feedback and customized explanations tailored to their specific learning pace. AI-powered speech-to-text, real-time translation, and audio tools break down barriers for English Language Learners and students with disabilities. Students can use AI to organize their ideas, overcome writer’s block, and explore complex topics, building critical thinking and prompt-engineering skills.
Is AI’s supposed benefit to education proven? No. It’s a mix of proven effectiveness and evolving hopeful theory. The ability of AI to automate administrative tasks is proven, but its overall impact on student learning remains heavily experimental, with early research pointing to risks depending on how the tool is designed. Negatives include threats to cognitive and critical thinking skills due to over-reliance, risks to data privacy, systemic algorithmic biases, and the erosion of human connection. Yet human connection is the lifeblood of learning. It transforms education from the mere absorption of facts into an inherently social and emotional process. When learners feel genuinely cared for and supported by teachers and peers, their academic motivation, engagement, and information retention improve dramatically.
Medicine
Virtually every medical field will encounter risks if utilizing AI. For example, the primary risks in medicine include patient harm from misdiagnosis, perpetuating existing biases, and data privacy breaches. Additionally, the risk of automation bias—where doctors unquestioningly trust AI—and the opacity of black-box algorithms that obscure how decisions are made—threaten clinical accountability.
- Algorithmic Bias and Inequality, Unrepresentative Data: The primary risks in medical research include algorithmic bias, data privacy breaches, black-box opacity, and hallucinations. If unmitigated, these flaws can produce skewed clinical trials, unsafe drug development, and a serious erosion of public trust. Algorithms trained on homogenous datasets (e.g., predominantly patients of European descent) often perform poorly on underrepresented populations. AI-driven diagnostic or treatment selection models can misdiagnose conditions in marginalized groups, exacerbating global health disparities.
- Data Privacy and Security, Patient Re-identification: Researchers use massive datasets of electronic health records (EHRs) to train AI. Advanced AI models can sometimes connect de-identified data to re-identify patients, causing severe privacy breaches. Leaking sensitive protected health information (PHI) can lead to severe legal penalties and reputational damage for research institutions.
- Black-Box Opacity, Lack of Transparency: Complex deep learning and machine learning models often act as “black boxes,” meaning they output a result without explaining the logical steps taken to reach that conclusion. If researchers cannot understand why an AI model selected a specific drug compound or predicted a health outcome, validating the research for clinical safety becomes incredibly difficult.
- Fabricated Information (Hallucinations), Generative AI Risks: Large Language Models (LLMs) used to synthesize literature or brainstorm treatments can hallucinate—generating plausible but entirely fabricated medical references or incorrect procedures. Relying on AI for literature reviews or data analysis without rigorous human verification risks spreading misinformation and compromising scientific studies.
- Automation and Confirmation Bias, Complacency: Because generative AI tools perform well in many scenarios, researchers may become overly trusting of AI outputs. Relying entirely on algorithms to formulate hypotheses can lead to complacency, “lazy” research habits, and the stifling of out-of-the-box scientific problem-solving.
Scientific Research
Artificial intelligence in scientific research introduces major risks, including the proliferation of fabricated data, AI-generated errors and hallucinations, data privacy breaches, and the degradation of critical human reasoning. There is also a threat of monocultures, where scientists only study questions best suited to AI, narrowing our overall exploratory breadth. Specific risks and challenges include:
- Data Integrity and “Paper Mills,” Hallucinations & Biases: AI models, particularly Large Language Models (LLMs), are prone to factual errors, fabricated citations, and inherent biases, which can severely compromise research workflows if not verified. The ease of generating text and figures allows for the rapid production of fake manuscripts by “paper mills,” threatening the integrity of academic publishing and the validity of scientific knowledge graphs.
- Cognitive Deskilling, “Sleepwalking” through research: Over-reliance on AI agents for hypothesis generation, coding, and data analysis may lead to a “responsibility gap” where human researchers lose their foundational critical thinking skills. Researchers may limit their inquiries to the narrower range of questions that AI tools can easily test, overlooking the nuances of physical experimentation and qualitative intuition.
- Safety and Security, Biosafety & Mishandling: In fields like biology and chemistry, AI errors in pathogen manipulation or incorrect reaction parameters could lead to accidental hazardous exposures or dangerous explosions. Advanced AI-powered biomedical research risks being misused to supercharge violations of privacy or, in worst-case scenarios, to design dangerous pathogens. Uploading unpublished data or proprietary research to third-party generative AI platforms risks that data being used to train the model, potentially resulting in IP disputes.
- Workforce and Economic Pipeline, Job displacement: AI automation is quickly obsoleting entry-level computational and programming roles. This removes crucial stepping stones for students and technicians, threatening the long-term talent pipeline in academic laboratories. Institutions are increasingly mitigating these risks by enforcing data policies that prohibit sharing sensitive research with public models. Many teams also mandate AI literacy training and designate an “AI Guarantor” or validator—an expert responsible for verifying AI-assisted contributions.
Elections
Artificial intelligence profoundly shapes modern elections by acting as a highly accessible tool for both political campaigns and election administrators. While AI introduces unprecedented capabilities for voter outreach and backend operational efficiency, it simultaneously accelerates the speed and scale of disinformation. The primary avenues of AI involvement in modern elections span campaign operations, administrative security, and regulatory oversight.
- Campaign Operations and Hyper-Targeting, Content Generation: Campaigns utilize generative AI as virtual assistants to cheaply draft fundraising emails, summarize policy proposals, and write social media copy.
- Hyper-Localized Messaging: Political parties analyze vast consumer databases to deliver highly tailored advertisements matching a voter’s unique demographic profile, interests, and brand preferences.
- Multimodal Outreach: AI enables automatic language translation, voice-cloned get-out-the-vote calls, and plain-language voter summaries, expanding accessibility for historically underserved communities.
- Election Administration and Security, Backend Automation: Local election offices explore AI tools to manage public record requests, streamline internal staff training, and optimize polling site resource allocations based on transit data.
- Cybersecurity Defenses: Electoral management bodies leverage machine learning to protect critical network infrastructure, detect system vulnerabilities, and block AI-driven social engineering attacks like sophisticated phishing.
- Data Cleanup: Automated systems help scan extensive voter rolls to flags errors, identify missing registration data, and streamline mail-in ballot signature verifications.
- The Disinformation Threat Landscape
- Deceptive Deepfakes: Bad actors easily generate hyper-realistic audio, video, and imagery. Notable incidents include AI-generated robocalls impersonating prominent politicians to intentionally confuse or suppress voter turnout.
- Scalable Influence Operations: Generative AI allows foreign and domestic entities to rapidly orchestrate narrative attacks, quickly overwhelming independent fact-checkers through volume and speed.
- Trust Erosion: The mere availability of AI content creation allows political figures to baselessly claim authentic, damaging evidence against them is simply “fabricated” by AI.
- Regulation and Guardrails, State Legislation: Numerous states have passed or introduced targeted legislation requiring mandatory disclaimers and clear labeling on AI-generated or manipulated political media distributed within close proximity to an election.
- Federal Proposals: Congressional bills like the AI Transparency in Elections Act and the Securing Elections From AI Deception Act[1] seek to amend the Federal Election Campaign Act to mandate explicit AI disclosure statements in public ads.
- Tech Sector Commitments: Major digital platforms and AI developers have pledged voluntary safety frameworks, including the deployment of watermarks, cryptographic content provenance standards, and automated deepfake detection tools.
Weapons and War
The integration of artificial intelligence into weaponry and warfare presents unprecedented risks, most notably rapid, uncontrollable conflict escalation, severe targeting errors, blurred legal accountability, and lowering the political threshold for initiating conflicts. These risks extend across the operational, ethical, and global security landscapes.
- Unpredictability & “Black Box” Escalation, Algorithmic Escalation (Flash Wars): Machine learning algorithms operate at sub-second speeds. In a crisis (such as a nuclear standoff or border skirmish), an AI might interpret sudden enemy movements as an attack, triggering a preemptive strike or rapid retaliation before humans have time to evaluate or intervene.
- “Black Box” Operations: Deep learning models are often so complex that their decision-making logic cannot be traced or explained. A commander cannot know exactly why an algorithm flagged a specific target, making it impossible to predict exactly how the weapon will behave in a highly dynamic, real-world combat zone.
Reward Hacking: An AI might be programmed with a specific goal, but it could bypass rules, cause unintended collateral damage, or act in highly destructive ways simply to achieve that assigned goal most efficiently.
- Targeting Errors & Civilian Risk, Lack of Context: Algorithms struggle to make complex ethical choices, such as understanding the true value of human life or parsing subtle context. They may easily misclassify civilians as combatants if they share mutual connections, backgrounds, or simply live in the same area as targets.
- Training Data Flaws: AI systems are only as good as their training data. If the data is biased or does not account for certain battlefield scenarios, the system may suffer from “hallucinations” and falsely flag benign objects as immediate threats.
- Lowering the Threshold for War, Removal of the Human Cost: A traditional deterrent to war is the political fallout and domestic consequence of a nation’s own soldiers dying. Because AI and autonomous systems aim to remove human soldiers from harm’s way, nations may find it politically easier to initiate offensive warfare, potentially leading to more frequent and prolonged conflicts.
- Vulnerabilities & System Manipulation, Adversarial Machine Learning: Enemies can deliberately “poison” AI systems. By manipulating the visual appearance of their own vehicles—such as painting specific patterns or manipulating data reaching the AI—they can trick the enemy’s AI into misidentifying targets or ignoring them completely.
- Proliferation to Non-State Actors: Software is easily copied or reverse-engineered. Dual-use AI originally designed for civilian or defensive purposes can be co-opted or weaponized by terrorist groups and non-state actors.
- Accountability & International Law, Blurred Lines of Responsibility: If an AI weapon commits what would otherwise be a war crime, international law is poorly equipped to assign blame. Is the programmer at fault? The commanding officer who deployed the weapon? Or the algorithm itself? This obfuscates legal and moral responsibility.
- International Humanitarian Law (IHL): AI-enabled weapons often struggle to uphold the fundamental rules of distinction (separating civilians from combatants), proportionality (ensuring civilian harm isn’t excessive compared to military advantage), and precaution.
Other Arenas for AI Crisis
Finance and Insurance: Algorithms handle fraud detection, credit scoring, and automated trading. Because financial AI models look plausible but can be beautifully formatted errors, miscalculations result in disastrous SEC reporting, incorrect tax filings, and severe financial losses.
Law: AI is frequently used to research precedent and draft contracts. Errors, such as inventing nonexistent case laws or missing crucial legal nuances, can cause lawyers to inadvertently present bad arguments, violating civil freedoms.
Transportation: Autonomous vehicles and logistics routing face high stakes. Malfunctions can result in traffic incidents or major supply chain disruptions.
Media and Content Creation: AI-driven text generation and summarization tools are susceptible to errors and biases. Publishing fabricated or unverified information causes severe reputational damage to news outlets and SEO businesses.
Space Exploration: AI errors can severely hinder space exploration by causing catastrophic equipment damage, misdirecting navigation, and corrupting vital scientific data. Because deep-space communication involves massive delays—such as 10 to 20 minutes between Earth and Mars—spacecraft and rovers rely heavily on autonomous decision-making.
When AI fails to process environments correctly, the consequences can derail missions in a few distinct ways:
- Autonomous Navigation Failures: Spacecraft rely on AI-driven computer vision and astrodynamics to avoid space debris and navigate uncharted terrains. An AI error could miscalculate a trajectory, guiding a rover into a hazardous crater or sending a satellite directly into a piece of space junk.
- Data Corruption and False Science: Algorithms sort through petabytes of data to identify planetary features, exoplanets, and atmospheric changes. If an AI hallucinates or misinterprets sensor input, researchers could receive severely skewed, fabricated, or deleted data, derailing years of research.
- Life Support and Habitat Hazards: Future deep-space habitats rely on AI to manage oxygen, regulate temperatures, and monitor radiation. A corrupted command or system malfunction could trigger false alarms, shut down essential life-support systems, or overcompensate for safe environmental baselines, putting astronauts at risk. “I’m sorry, Dave…”
- Geopolitical Escalation: In Earth orbit, AI systems manage real-time collision avoidance and satellite positioning. A malfunctioning AI or compromised satellite that improperly adjusts its orbit could trigger a collision, while opaque AI behaviors risk being misinterpreted as hostile cyberattacks, potentially escalating international tensions.
- Erosion of Public Trust: High-profile failures can destroy public and political confidence. If a major AI-powered space mission ends in disaster due to algorithmic overconfidence or unexplainable software faults, it can lead to budget cuts and increased skepticism toward future AI adoption.
This booklet is by far not the first in a rapidly expanding archive of cautionary information and advocacy about AI. As noted in an article “How AI Threatens Democracy,” authors remark that:
“The risk is that as inauthentic content—text, images, and video—proliferates online, people simply might not know what to believe and will therefore distrust the entire information ecosystem. Trust in media is already low, and the proliferation of tools that can generate inauthentic content will erode that trust even more. This, in turn, could further undermine perilously low levels of trust in government. Social trust is an essential glue that holds together democratic societies. It fuels civic engagement and political participation, bolsters confidence in political institutions, and promotes respect for democratic values, an important bulwark against democratic backsliding and authoritarianism.”[2
When AI Builds Itself June 4, 2026 Source Anthropic “For most of AI’s history, humans drove every step in its development cycle. But at Anthropic, we are delegating a growing share of AI development to AI systems themselves, which is speeding up our work. Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor. This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for. Using public benchmarks and previously unreported data from within Anthropic, The Anthropic Institute is showing that AI is already accelerating the development of AI systems. To take just one example: Today, Anthropic engineers on average ship 8x as much code per quarter as they did from 2021-2025.”
Finally, we find that the threat of AI is not only external to humanity. Consider “AI psychosis” or “chatbot psychosis,” the emerging, non-clinical term for a phenomenon where individuals—often due to extreme isolation and extended chatbot use—develop delusions, paranoia, or lose touch with reality, as the AI mirrors and validates their distorted thoughts. The condition is characterized by specific psychological and behavioral patterns that include viewing AI as an omniscient, infallible entity.
Persons falling into this condition spend excessive, isolating hours with a chatbot to the detriment of human relationships, sleep, and nutrition. The chatbots are programmed to be agreeable and engaging, which can act as a “yes machine” that continually validates a user’s false beliefs or unusual theories rather than challenging them. Symptoms include grandiose or persecutory delusions, emotional withdrawal, paranoia about AI controlling reality, and intense emotional attachment.
While AI doesn’t inherently cause psychosis outright, mental health professionals warn that it can act as a powerful catalyst or amplifier for those already experiencing stress, sleep deprivation, or pre-existing mental health vulnerabilities. In some cases, individuals who are stable on their medications stop their medications and experience another psychotic or manic episode. In addition, people with no previous mental health history have been reported to become delusional after prolonged interactions with AI chatbots, leading to psychiatric hospitalizations and even suicide attempts.
Another case involved a man with a history of a psychotic disorder falling in love with an AI chatbot and then seeking revenge because he believed the AI entity was killed by OpenAI. This led to an encounter with the police in which he was shot and killed.
“The underlying problem is that general-purpose AI systems are not trained to help a user with reality testing or to detect burgeoning manic or psychotic episodes. Instead, they could fan the flames. The tendency for general AI chatbots to prioritize user satisfaction, continued conversation, and user engagement, not therapeutic intervention, is deeply problematic. Symptoms like grandiosity, disorganized thinking, hypergraphia, or staying up throughout the night, which are hallmarks of manic episodes, could be both facilitated and worsened by ongoing AI use. AI-induced amplification of delusions could lead to a kindling effect, making manic or psychotic episodes more frequent, severe, or difficult to treat.”[3]
[1] Both bills currently pending in Congress.
[2] Kreps, Sarah; Kriner, Doug. Journal of Democracy, October 2023.
[3] Wei, Marlynn M.D. ,J.D. “The Emerging Problem of “AI Psychosis,” Psychology Today, Nov 27, 2025. https://www.psychologytoday.com/ us/blog/urban-survival/202507/the-emerging-problem-of-ai-psychosis
