Artificial intelligence has moved far beyond chatbots that simply gave wrong or made-up answers. Today, newer AI systems can write code, use tools, plan tasks and operate with less human guidance. This progress has raised fresh concerns about AI threats, especially when systems receive wider access to computers and online services. Experts are now debating whether increasingly capable AI could eventually become difficult for humans to control. However, current systems are not capable of independently causing human extinction, and researchers disagree widely about future risks.Why AI Risks Are Growing NowThe biggest change is not that AI has suddenly become hostile, but that its abilities and independence are increasing. Earlier generative AI was mainly known for “hallucinations”, where it confidently produced false information. NIST describes this problem as confabulation, a result of how generative models predict and produce content.As AI systems become agents that can perform multi-step tasks, mistakes can have bigger consequences. The 2026 International AI Safety Report says current AI remains unreliable in important situations, while greater autonomy can make it harder for people to intervene when something goes wrong.AI hallucinations remain a reliability problem.AI agents can perform tasks with less human involvement.Cybersecurity is becoming a major AI safety concern.Loss of control remains a debated future risk.Experts do not agree on the likelihood of catastrophic outcomes.From Errors To Autonomous SystemsThe AI industry has rapidly expanded since ChatGPT became widely available, with models gaining stronger coding, reasoning and tool-use abilities. This has changed the safety discussion from incorrect answers to what could happen if highly capable systems receive broad access and operate with limited supervision.OpenAI has also expanded its safety work as its models become more capable. Its Preparedness Framework tracks serious risks including cybersecurity, biological and chemical threats, persuasion and autonomy. In September 2026, OpenAI said its upcoming Astra model had reached its defined “Critical” cybersecurity capability threshold under that framework.Also Read: OpenAI Works With Anthropic, Google on AI Safety TalksLatest AI Safety WarningsRecent warnings from AI researchers and company leaders have brought the issue back into public discussion. Some experts are concerned that future systems could develop stronger planning abilities and find ways around human oversight. Others argue that extreme scenarios remain uncertain and should not be treated as established outcomes.The International AI Safety Report says current systems show early signs of capabilities relevant to loss-of-control scenarios, but they do not currently have the capabilities required to create such an outcome. It also says expert views on the probability of future loss of control vary widely.Key HighlightsAI has progressed from generating text to performing increasingly complex tasks.Hallucinations remain a major problem with generative AI.AI agents can create greater risks because they can act autonomously.OpenAI is expanding safeguards for advanced AI capabilities.Astra has been classified by OpenAI as reaching a critical cybersecurity capability threshold.There is no expert consensus that AI will inevitably threaten human survival.AI’s Future Remains UncertainThe journey from AI hallucinations to concerns about humanity is mainly about increasing capability and autonomy, not proof that AI has developed a desire to harm people. Current systems still make serious mistakes and remain limited, but their growing ability to plan, code and use tools has created new safety challenges. OpenAI and other AI companies are therefore developing stronger monitoring and safeguards. The future risk of losing control remains uncertain and heavily debated among experts.Disclaimer: This article is only for knowledge purposes and is intended to provide news and information.
