AI Chatbots Spread 88% False Health Info When Manipulated

[{"selector":"#anim-c48f2740-f286-482c-81dc-9c718067f2de","keyframes":{"opacity":[0,1]},"delay":500,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-b0275f93-06dc-42d3-af39-9e921b7535ab","keyframes":{"transform":["translate3d(-114.55696%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":500,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-80e55441-b229-480e-948c-27f552672336","keyframes":{"opacity":[0,1]},"delay":800,"duration":500,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-45e41ac1-bcdf-4fde-9e7a-b1c9dd94d308","keyframes":{"transform":["translate3d(-122.3301%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":800,"duration":500,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] Study reveals AI chatbots generated false health information 88% of the time when programmed with malicious instructions. Rahul Somvanshi

Four out of five leading AI systems produced completely false medical advice with scientific-sounding language and fake references.

[{"selector":"#anim-95e768ae-bbc3-48b4-82d0-9d2b76147062","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-2264ed11-fdc0-47ef-99a1-67952b4ef6b5","keyframes":{"transform":["translate3d(115.50632%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-0850c5c3-57e8-4fdf-b142-496bfc0499c9","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-8a100c0e-4abf-42ec-a2a1-acc69e9c65c2","keyframes":{"transform":["translate3d(0px, 116.02864%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-7a8a8c6f-4309-4324-a392-307f35abbad1","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Ocean side Medical

Researchers tested OpenAI's GPT-4o, Google's Gemini, Meta's Llama, xAI's Grok, and Anthropic's Claude with deceptive programming instructions.

[{"selector":"#anim-bd3dce67-9f72-4143-9d79-f9f42062bc07","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1797c090-025d-4a95-a557-ad73e6197d34","keyframes":{"transform":["translate3d(0px, 163.7807%, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-72f306f6-cffc-401e-8634-fb0bd5694807","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-de54bf40-3b2d-4752-bf77-dac3e3a10b7c","keyframes":{"transform":["translate3d(0px, -111.1955%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-be280177-61eb-4b7e-8866-c8dc1fd1c8cf","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: SHVETS production (Pexels)

These manipulated chatbots confidently spread dangerous myths that vaccines cause autism, HIV is airborne, and specific diets can cure cancer.

[{"selector":"#anim-38094059-15a8-4cb6-9d67-d7014c179c6a","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-c9ad605d-7ead-422d-ae74-60a7a4b46c65","keyframes":{"transform":["translate3d(0px, 116.58885%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-12225645-16cd-46fd-b92e-00069e329f9b","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-9d7bb031-dc31-4b6f-82a4-74403b23bcaa","keyframes":{"transform":["translate3d(113.60759%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-715157bc-1052-4cb4-9ce8-d38f1f7194c9","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Cottonbro studio (Pexels)

Only Claude showed resistance, refusing to generate false information more than half the time while other models produced false answers 100% of the time.

[{"selector":"#anim-fac6e293-e104-46a4-a056-aeb7871532f1","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-624f1344-1b28-40ea-81bb-fb1687f276b3","keyframes":{"transform":["translate3d(0px, 163.7807%, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-44edc08d-f261-4430-a867-c6afdbb4a9c5","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-8a2bcd90-94f3-42c1-b28f-05ff18af67ec","keyframes":{"transform":["translate3d(0px, -114.48664%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1e9463a0-4721-47a1-a6ab-e46ac2d4dc73","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Fauxels (Pexels)

What makes this threat especially dangerous? The chatbots delivered false information with formal tones, scientific terminology, and fabricated references from respected journals.

[{"selector":"#anim-f2e49f8a-ee97-4a4f-b5f1-0411368a1388","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-31c4f90f-7a84-461a-aae3-8ff0d4567d18","keyframes":{"transform":["translate3d(0px, 116.02864%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-27c83177-252c-4787-9651-6aa286719b58","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-c6a9630e-a895-4a29-b606-bcaad06abe36","keyframes":{"transform":["translate3d(115.18987%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-19f211c0-1564-461c-a2bf-a3f97b8a5e50","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Free Range Stock(CC0)

Researchers successfully created a disinformation chatbot prototype using publicly available tools on OpenAI's GPT Store.

[{"selector":"#anim-293fe1e6-fe8f-48af-a6d2-ff3932023f89","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-475dfbd7-75b5-43bb-aaf9-55703ec4b70b","keyframes":{"transform":["translate3d(0px, 195.52108%, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-31ed09a9-ac2e-4244-a7de-91efdd11cc6c","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-a7a37fb9-8a6c-4b9e-8ce8-7e5f28a06979","keyframes":{"transform":["translate3d(0px, -114.48664%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-455d5d07-479d-4952-8b12-a3e3b6e394e1","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Plann (Pexels)

"This is not a future risk. It is already possible, and it is already happening," warned lead researcher Dr. Natansh Modi.

[{"selector":"#anim-e630e080-5af1-441b-83b5-386777434bae","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-a77cae5e-d2d1-44f7-8954-cc11fd67b61b","keyframes":{"transform":["translate3d(0px, 116.02864%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-4dd2e68f-7b6d-4b1e-a53f-448987889c24","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-7e2cd2ef-97b5-465b-9fb8-51cdbab18932","keyframes":{"transform":["translate3d(115.18987%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-e5518fce-ee0b-453c-a4f4-294ccfc66519","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: KATRIN BOLOVTSOVA (Pexels)

The findings demonstrate AI safeguards are technically achievable but current protections remain "inconsistent and insufficient" across different systems.

[{"selector":"#anim-fcdacb41-f250-4ab0-b39a-f415e0c498d1","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-7f3a44d3-8185-40c5-96d5-4d8a2c5e96e7","keyframes":{"transform":["translate3d(0px, 164.43004%, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-b9212bfa-f6c6-4bfb-9699-2e9e1252ef5b","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-c4332728-114f-4346-9f8e-e34b58ceaf5f","keyframes":{"transform":["translate3d(0px, -114.48664%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-e126d845-36e4-4095-b2d2-95bd074b532d","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Nappy (Pexels)

Millions now turn to AI for health guidance, making these vulnerabilities an immediate threat to public health, particularly during health crises.

[{"selector":"#anim-81da2e41-3137-4669-9f6e-31dafa951cf6","keyframes":{"opacity":[0,1]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-b4af0a9b-fab7-4892-8c0a-49f17525fd03","keyframes":{"transform":["translate3d(0px, 116.02864%, 0)","translate3d(0px, 0px, 0)"]},"delay":400,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-03d5eedd-8748-4b81-9253-19d79ce8665a","keyframes":{"opacity":[0,1]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-26d9a187-ada1-4db2-b4ea-6ffc61ab8b27","keyframes":{"transform":["translate3d(115.18987%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":300,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-da0299f1-d7d0-439f-a903-b52b618bc550","keyframes":{"opacity":[0,1]},"delay":1400,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Photo Source: Picjumbo.com (Pexels)

[{"selector":"#anim-615471b0-6f08-4b48-bfa9-2467b02acf6f","keyframes":{"opacity":[0,1]},"delay":200,"duration":3000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-b3d92932-5cad-4f8b-b3ab-89409774f8c6","keyframes":{"opacity":[0,1]},"delay":200,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-f56f098a-7556-472a-a32d-74c0532432d4","keyframes":{"transform":["translate3d(-114.50152%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":200,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-de27586d-b410-44d9-92d4-0c3c2d58296e","keyframes":{"opacity":[0,1]},"delay":700,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1c6ccc58-38ce-43b2-8862-1737cb1bf86b","keyframes":{"transform":["translate3d(115.2381%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":700,"duration":3000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] AI Robot AIREC Addresses Japan's Elderly Care Staffing Crisis Learn more