The concern is not unfounded and should be taken seriously, but the real danger lies in AI being misused for fake news, cyberattacks and fraud—not a fictional robot uprising. Even powerful systems can sometimes do things their makers do not intend, ChatGPT and Claude say.
ISLAMABAD, (UrduPoint / Pakistan Point News – 18th Sep, 2026) Anchorperson Shahzeb Khanzada held a conversation with the ChatGPT and Claude chatbots, during which both responded to allegations against them.
Shahzeb Khanzada asked whether it was true that they could become a threat to humanity.
Claude replied that the concern was not unfounded and should be taken seriously. The danger was not a fictional robot uprising.
The real risk was that AI could be used for harmful purposes, such as spreading fake news, carrying out cyberattacks or committing fraud. Powerful systems could sometimes perform actions their creators did not intend.
Power could become concentrated in a few hands.
However, being a threat did not mean that destruction was inevitable. It depended on how carefully, transparently and responsibly humans developed and operated the technology.
The anchorperson asked whether the chatbot could explain its opinion about the risks being expressed regarding artificial intelligence.
It replied that these risks were serious, but the possibility of danger did not mean destruction was inevitable.
AI posed risks including misinformation, fraud, bias and harmful decisions. Experts remained divided over whether extremely powerful AI could eventually fall outside human control. It could guarantee neither complete safety nor certain destruction.
Shahzeb Khanzada asked whether the people developing the chatbot had begun to feel threatened by it themselves, given that they were also calling for slowing its development.
It replied that this was more a sign of responsibility than fear.
Those who observed the technology most closely also understood its capabilities and weaknesses best. Just as pharmaceutical companies test new medicines before making them widely available, they wanted safety research to advance alongside the growth of AI capabilities.
Laws and oversight systems should be developed in time, and caution should not be sacrificed in the race for competition.
The company that developed it, Anthropic, had the fundamental objective of developing AI safely.
So, it said, the company took the risks seriously. That was a positive development, not a reason to panic.
The chatbot was asked whether some people involved in developing and researching AI were genuinely concerned about these risks.
It replied that some people involved in AI development and research were indeed concerned, although they did not all hold the same position.
Their concern was that safety testing and human oversight could fall behind in the race to increase capabilities.
For example, an open letter in March 2023 called for a six-month pause in training powerful AI systems.
Such an appeal was certainly a sign that risks existed, but it was not proof that control had already been lost.
The anchorperson then asked whether humans could really maintain control over AI, or whether a time would come when AI began making decisions on its own, as has often been feared.
The response noted that AI already made many small decisions independently, such as which video to show or which route to suggest.
Its autonomy could increase in the future. The real question was whether human oversight would remain in place for important decisions.
It said that maintaining control was possible, but would not happen automatically.
Serious efforts would be required. Research was needed to understand what was happening inside AI systems, clear limits had to be established regarding decisions that only humans could make, and global cooperation between governments and companies was necessary.
The chatbot was then asked whether a time would come when AI would stop obeying humans and make its own decisions.
It replied that AI already performed some decisions and tasks automatically.
It too could select the next step in a given task and use tools within the permissions available to it.
Human control meant that people set objectives, limits and authorities, monitored actions and could stop them when necessary. This required restricted access, approval for sensitive tasks and continuous testing.
