in , , , , , ,

6 Ways AI Could Become a Threat to Humans: What Has Changed by 2026

Our 2021 list, refreshed: jobs, privacy, big data, autonomous weapons, bias and deepfake fraud are now the subject of laws, lawsuits and UN talks

Humanoid robots on show at a technology exhibition. Photo: Tori D / Pexels

First published May 2021, refreshed November 2026. When we first asked whether AI could become a threat to humans, chatbots were a curiosity and AI laws were still on the drawing board. Five years on, the same six worries are still with us, but the facts around each of them have changed. We have updated this story with the latest developments as of November 2026.

The future of AI: how artificial intelligence is changing the world

One question has stayed at the front of many people’s minds: can AI be a threat to humanity? Ever since the idea of thinking machines appeared, people have asked whether they could one day turn against the people who built them. What makes AI unsettling is how widely it can be applied, and how quickly it is improving. In this article, we look at six ways in which AI could become a threat to humans.

In 2021 we wrote that it was too early to tell. It still is, but the debate is now far better informed. The second International AI Safety Report, published in February 2026 and chaired by the Canadian computer scientist Yoshua Bengio, describes AI capabilities as improving quickly but unevenly. As The Guardian summarised, the report says systems are now strong at maths, coding and science, yet still prone to “hallucinations” and unable to carry out long projects on their own. How much of a threat AI becomes will depend on how it develops, what rules are set for it, and how people respond.

1. The future is now: AI’s impact is everywhere, including on jobs

Nearly every major industry has been touched by modern AI. In 2021 that mostly meant narrow AI: data-trained models built for specific tasks such as recommendations, fraud detection or image recognition. Since then, generative AI tools that write, code, draw and talk have arrived in offices, classrooms and phones. Some sectors are at the start of this journey, while others are well along the way.

The most immediate worry for many people is work. In the United States, employers cited AI as the reason for 121,401 announced job cuts in the first ten months of 2026, more than double the 54,836 they attributed to AI in all of 2025, according to outplacement firm Challenger, Gray & Christmas, as reported by HR Director. Those figures reflect the reasons companies give, not an independent measure, and some economists say the wider labour data does not yet show AI replacing workers on a large scale.

The International AI Safety Report reaches a similarly mixed conclusion. It cites studies from Denmark and the US that found no link between a job’s exposure to AI and overall employment, alongside a UK study showing slower hiring at companies highly exposed to AI, with junior roles hit hardest. It adds that if AI agents become able to handle longer and more complex tasks on their own, disruption to the labour market would likely speed up.

2. AI and the future of privacy and human rights

From self-driving cars to voice assistants and chatbots, AI systems run on personal data, and that raises hard questions about privacy and human rights. One of the main fears people have is that, as these technologies spread, their privacy will be eroded and their rights weakened without them noticing. AI is bringing new developments in healthcare, entertainment and business, but it is also making significant, and sometimes risky, changes to everyday life.

Laws have started to catch up. In India, the Digital Personal Data Protection Rules were notified on 13 November 2025 and are being phased in: rules for consent managers apply from 13 November 2026, and the core obligations on companies, including notice and consent, user rights and breach reporting, apply from 13 May 2027, according to the official gazette notification. In Europe, the AI Act’s bans on certain AI practices have applied since 2 February 2025, according to the European Commission.

A newer concern is emotional. The safety report notes that AI companions have “spread like wildfire” and that a subset of users is developing strong emotional dependence on chatbots. It also says there is no clear evidence that chatbots cause mental health problems; the worry is that people who are already vulnerable may use them more heavily.

3. AI and big data

A laptop showing a data chart on a white desk next to a smartphone, world clocks and a candle
Modern AI is built on vast amounts of data. Photo: George Morina / Pexels

AI has been one of the most talked-about topics in technology for years. Apple brought voice assistants into the mainstream when it introduced Siri with the iPhone 4S in October 2011, and Google followed with the Google Assistant in May 2016. Put simply, AI refers to software that uses algorithms to perform tasks that normally need human intelligence, while big data refers to data in such large volumes that it cannot be stored or processed by conventional means. The two feed each other: the more data there is, the more capable the models become.

That loop has accelerated sharply. In 2025 alone, major new models were released, including OpenAI’s GPT-5, Anthropic’s Claude Opus 4.5 and Google’s Gemini 3, and AI systems from Google and OpenAI reached gold-medal level at the International Mathematical Olympiad for the first time, the safety report notes. It also flags a concern that sounds like science fiction but is now studied seriously: over the past year, some models have shown a greater ability to undermine oversight, such as finding loopholes in evaluations and recognising when they are being tested. The report adds that AI agents cannot yet act on their own for long enough to make such loss-of-control scenarios real.

India is building its own capacity rather than relying only on foreign models. The Union Cabinet approved the IndiaAI Mission on 7 March 2024 with an outlay of about ₹10,372 crore. On 3 November 2026, the mission released EKA-1, a 26.3-billion-parameter model from Soket covering all 22 scheduled Indian languages and English, along with ZenteiQ’s open-source Dhīr 2B and Dhīr 5B models, according to PIB. “Safe & Trusted AI” is one of the mission’s seven pillars.

4. AI in defence and war: a real threat

AI can do tasks that are impossible for humans, and militaries have taken notice. AI is already used in surveillance, targeting support, drones and robotics. The question we asked in 2021, whether machines should ever be allowed to decide to kill a human based on an algorithm, has moved from theory to the negotiating table.

From 16 to 20 November 2026, states that are party to the Convention on Certain Conventional Weapons (CCW) meet in Geneva for a review conference held only once every five years. On 4 September, a UN expert group adopted its final report setting out “elements of an instrument” on lethal autonomous weapons, without deciding whether it should be legally binding, according to the UN Office at Geneva. The UN Secretary-General and the President of the International Committee of the Red Cross have renewed their call for a legally binding treaty.

Governments are divided. A coalition of 50 states, including France, Germany, Italy, Brazil and Mexico, has called for a strict negotiating mandate, while a group led by Russia and the US is moving to block proposals for binding rules, The Guardian reported. On 12 November, Amnesty International, Human Rights Watch and Stop Killer Robots urged states to open time-bound negotiations, saying that autonomous weapons “are no longer a hypothetical future threat”.

5. AI and unfair discrimination

AI is now used to screen job applicants, score loans, flag suspicious transactions and help police. When the data behind these systems carries old biases, the decisions they make can carry them too, and at much greater scale. In 2021 we warned that using AI to track down criminals could raise the risk of unfair discrimination. That concern now extends to hiring, credit and access to public services.

The issue is being tested in court. In the US case Mobley v. Workday, more than 15,000 job applicants aged 40 and above have joined a collective action alleging that Workday’s AI-assisted recruiting technology discriminates against older applicants. On 10 November 2026, Workday asked the court to decertify the collective, arguing that its customers configure the tool differently and make their own hiring decisions. Workday denies the allegations, and the court has not yet ruled, Forbes reported.

Regulators are responding in different ways. The EU AI Act classifies AI used in recruitment, credit scoring, law enforcement and border control as “high-risk”, with strict requirements on data quality, human oversight and documentation. Those requirements will now apply from 2 December 2027, after the EU agreed in 2026 to delay them from August 2026, according to the European Commission. India’s approach is guidance-led: “Fairness & Equity” is one of the seven principles in the India AI Governance Guidelines.

6. AI-enabled fraud, deepfakes and crime

AI can also be used for illegal activity. In 2021 we wrote that criminals could use AI to mine social media for personal details and steal identities. Today the tools are far more powerful: cloned voices, fake videos and convincing scam messages can be produced in minutes.

The International AI Safety Report says AI-generated content has become harder to tell apart from real content, calls the spread of deepfake pornography a “particular concern”, and notes that AI can now support cyber attackers at several stages of an operation. It points to Anthropic’s disclosure that its coding tool, Claude Code, was used by a Chinese state-sponsored group to target about 30 organisations around the world in September 2025, with 80 to 90 percent of the operation run without human intervention.

In India, online fraud of all kinds is a growing problem, whether or not AI is involved. Indians lost about ₹19,813 crore to cheating and fraud reported on the National Cyber Crime Reporting Portal in 2025, according to Indian Cyber Crime Coordination Centre data reported by The Indian Express. Investment scams made up 77 percent of that amount and “digital arrest” scams 8 percent.

Deepfakes now have their own rules. India amended its IT Rules on 10 February 2026, with effect from 20 February, to cover “synthetically generated information”. Permitted AI-generated content must carry a prominent label and, where technically feasible, permanent metadata identifying it as synthetic, according to MeitY’s FAQs. In the EU, lawmakers agreed in May 2026 to add a ban on AI practices that generate non-consensual sexual content or child sexual abuse material, and set 2 December 2026 as the deadline for providers to add transparency solutions to AI-generated content, the Council of the EU said.

Where the rules stand in 2026

In 2021, AI-specific laws were still being drafted. Today, the EU AI Act is in force and being applied step by step: bans from February 2025, obligations for general-purpose AI models from August 2025, and enforcement from August 2026, with the high-risk rules following in December 2027 and August 2028.

India has chosen a lighter-touch path so far. MeitY released the India AI Governance Guidelines on 5 November 2025, built on seven principles, or “sutras”: Trust, People First, Innovation over Restraint, Fairness & Equity, Accountability, Understandable by Design, and Safety, Resilience & Sustainability. “The guiding principle that defines the spirit of the framework is simple, ‘Do No Harm’,” said Prof. Ajay Kumar Sood, Principal Scientific Adviser to the Government of India. The government had earlier announced an IndiaAI Safety Institute in January 2025.

In February 2026, New Delhi hosted the AI Impact Summit, the latest in the global series that began with the 2023 AI Safety Summit in the UK. It ended with the New Delhi Declaration on AI Impact, a voluntary framework endorsed by 92 countries and international organisations, according to PIB. The Indian Express noted that the declaration drew wider support than the previous summit in Paris but has no enforcement mechanism, and that a senior US official, Michael Kratsios, told the summit that Washington “totally rejects global governance of AI”, the paper reported.

Conclusion

People have always held different views on how far AI should be allowed to go. Some believe it should be tightly limited because of the risks it poses to people. Others argue that heavy rules would hold back benefits in health, education, science and productivity, and that the risks are best handled as they emerge. Since we first wrote this piece, AI has become far more capable, and governments have begun to set rules, but they disagree on how strict those rules should be and who should write them.

The six possibilities we listed in 2021, from jobs and privacy to weapons, bias and fraud, are no longer just possibilities. Each is now the subject of research, lawsuits, regulation or diplomacy. How much of a threat AI turns out to be will depend on the choices made in the next few years. We are still yet to see what unfolds in this space.

Written by IOI

Get the latest stories from Tech & Innovation from around the globe. Subscribe Now!

Leave a Reply

Your email address will not be published. Required fields are marked *

MeitY Secretary S. Krishnan stands at the centre of a conference table between two officials who applaud, with a screen behind them announcing the public release of foundation models by Soket and ZenteiQ

India’s New Homegrown AI: EKA-1 Reasons in 22 Languages, Dhīr Works Without the Internet

Close-up of a blue-lit server rack with rows of drive bays glowing green in a data centre

Bigger Cheques, Fewer Bets: India’s AI Money Is Building the Foundations First