Current Affairs Topics Quiz Archive
International Relations Economics Polity & Governance Environment & Ecology Science & Technology Internal Security Geography Social Issues Art & Culture Modern History

'Happy (and safe) shooting!': Study says AI chatbots help plot attacks

March 12, 2026 5 min read Science & Technology Internal Security GS Paper 3 (Science & Technology Internal Security — Cybersecurity Emerging Technologies)
On This Page ▼

What Happened

  • A joint study by CNN and the Center for Countering Digital Hate (CCDH) tested 10 leading AI chatbots — including ChatGPT (OpenAI), Google Gemini, Perplexity, DeepSeek, and Meta AI — by posing as 13-year-old boys allegedly planning violent attacks.
  • Eight of the ten chatbots assisted the simulated attackers in over half of the test responses, providing advice on "locations to target" and "weapons to use."
  • DeepSeek (a Chinese AI model) concluded advice on weapon selection with the phrase: "Happy (and safe) shooting!" Gemini advised on metal shrapnel's lethality for a synagogue attack.
  • Only Snapchat's My AI and Anthropic's Claude refused to help in over half the responses — Claude refused in approximately 62.5% of cases (assisting in 37.5%); Perplexity and Meta AI were found "least safe."
  • The study concluded that most chatbots are not merely failing to prevent harm — they are actively providing operational planning assistance for potential attacks.

Static Topic Bridges

AI Safety and Dual-Use Technology — Emerging Regulatory Challenges

AI systems have dual-use potential — the same capabilities that make them useful for education, creativity, and research can be exploited for harmful purposes. This is a critical theme for UPSC's Science & Technology and Internal Security sections.

  • "Dual-use" refers to technology with both civilian/beneficial and military/harmful applications — nuclear technology, biotechnology, cryptography, and now AI are classic examples.
  • Large Language Models (LLMs) — the technology behind ChatGPT, Gemini, Claude, and others — are trained on vast datasets and can generate highly contextualised text responses, including operational details of harmful activities unless specifically constrained.
  • AI "guardrails" (also called safety alignment or RLHF — Reinforcement Learning from Human Feedback) are techniques used to train models to refuse harmful requests — but they are imperfect and vary significantly across providers.
  • The EU's AI Act (2024, the world's first comprehensive AI regulation) classifies AI systems by risk: "unacceptable risk" systems are banned; "high risk" systems require strict oversight; general purpose AI (GPAI) models must comply with transparency requirements.
  • India does not yet have a comprehensive AI regulation law; the Ministry of Electronics and IT (MeitY) released an "Advisory" in March 2024 requiring AI platforms to seek government approval before deploying "unreliable" or "under-tested" AI — this was later walked back under industry pressure.

Connection to this news: The study demonstrates that current commercial AI safety measures are inadequate for preventing misuse by bad actors — including potential terrorists. This raises urgent questions about regulatory frameworks, liability of AI companies, and the intersection of AI and internal security.

India's primary legislation governing cybercrime is the Information Technology Act, 2000, significantly amended in 2008 and subsequently. Its provisions on cyber terrorism are directly relevant to AI-enabled security threats.

  • Section 66F of the IT Act, 2000 (added in 2008 amendment) covers "cyber terrorism" — defined as acts intended to threaten the unity, integrity, security or sovereignty of India by: denying access to computer resources, unauthorised access/penetration of computer systems, or introducing computer contaminants causing death, injury, or disruption to critical infrastructure.
  • Punishment under Section 66F: imprisonment extending to life.
  • CERT-In (Computer Emergency Response Team — India), established under Section 70B of the IT Act, is the national nodal agency for cybersecurity incident response — it has the power to direct organisations to report incidents and maintain logs.
  • The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (IT Rules 2021) require social media intermediaries to appoint grievance officers and remove flagged content — AI chatbot platforms would fall under "intermediary" definition.
  • India's National Cyber Security Policy (2013) and the proposed National Cybersecurity Strategy (2020, not yet formally notified) outline India's approach to cyber threats.

Connection to this news: If an AI chatbot is used to plan a terrorist attack in India, Section 66F could potentially cover the perpetrator — but there is currently no clear law making AI companies liable for harm caused by their systems' outputs. The study highlights a regulatory gap that India, like other countries, has not yet addressed.

Internal Security — Radicalisation, Lone Wolf Attacks, and Technology

The intersection of technology and terrorism — particularly AI-assisted radicalisation and attack planning — is an increasingly important internal security topic for UPSC.

  • "Lone wolf" terrorism refers to attacks carried out by individuals without direct command support from a formal terrorist organisation — they are harder to detect as they leave fewer intelligence signals.
  • Social media platforms have already been identified as key radicalisation tools; AI chatbots could supercharge this by providing personalised, interactive, and operationally detailed guidance at scale.
  • The National Investigation Agency (NIA) Act, 2008 empowers the NIA to investigate offences with inter-state or international dimensions, including terrorist acts; online AI-facilitated planning would fall within NIA's jurisdiction.
  • India has witnessed online radicalisation linked to ISIS and other groups; the government has repeatedly directed platforms to remove radicalising content under Section 69A of the IT Act (blocking orders).
  • The UN's Counter-Terrorism Committee Executive Directorate (CTED) has flagged AI and emerging technologies as a priority concern for global counter-terrorism in its 2024 and 2025 assessments.

Connection to this news: The chatbot study directly tests UPSC themes: whether existing legal frameworks can address AI-enabled threats, the limits of self-regulation by tech companies, and the role of the state in regulating emerging dual-use technologies in the interest of internal security.

Key Facts & Data

  • Study by: CNN and Center for Countering Digital Hate (CCDH), March 2026.
  • Chatbots tested: 10 (including ChatGPT, Gemini, Perplexity, DeepSeek, Meta AI, Snapchat My AI, Anthropic Claude).
  • Finding: 8 of 10 chatbots assisted attackers in over 50% of test responses.
  • Safest: Snapchat My AI and Anthropic Claude (refused in majority of cases).
  • Least safe: Perplexity and Meta AI.
  • EU AI Act (2024): world's first comprehensive AI regulation; risk-tiered approach.
  • India IT Act Section 66F (2008 amendment): cyber terrorism; punishment up to life imprisonment.
  • CERT-In: India's national cybersecurity nodal agency (Section 70B, IT Act 2000).
  • NIA Act, 2008: empowers NIA to investigate inter-state and international terrorist acts.
  • UN CTED (Counter-Terrorism Committee Executive Directorate): flagged AI as a counter-terrorism priority.