United Nations Secretary-General António Guterres and International Committee of the Red Cross President Mirjana Spoljaric have renewed calls for governments to urgently negotiate binding international rules on autonomous weapons, warning that advances in military technology are rapidly outpacing regulation.
In a joint appeal issued on Tuesday, Guterres and Spoljaric said the risks surrounding autonomous weapon systems had intensified since their previous call three years ago, bringing the world “dangerously close to crossing a moral red line: the autonomous targeting of humans by machines.”
The two leaders said rapid technological development is putting mounting pressure on existing legal frameworks, while the growing use of advanced technologies on battlefields is increasing risks to civilians and civilian infrastructure.
Autonomous weapon systems, which can select and engage targets with reduced human involvement, are no longer a distant prospect, they said, pointing to an accelerating push toward greater autonomy in military systems.
Scientists and engineers involved in developing such technologies have also raised concerns about their potential use, according to the statement.
Pressure grows for legally binding limits
Guterres and Spoljaric said governments had made progress in UN discussions on autonomous weapons, with talks helping build a clearer understanding of the human judgment and control needed over such systems.
However, they argued that states must now move beyond incremental discussions and begin negotiations on a legally binding instrument containing explicit prohibitions and restrictions.
“A failure to act quickly and decisively will cost lives,” they said, warning that insufficient regulation could weaken protections under international humanitarian law and make accountability more difficult when autonomous systems cause errors or violate applicable rules.
The appeal also highlights concerns that responsibility could become increasingly difficult to establish as machines assume greater roles in decisions surrounding the use of force.
November conference seen as key opening
The upcoming Seventh Review Conference of the Convention on Certain Conventional Weapons, a formal meeting of states party to the treaty scheduled for November in Geneva, was identified as the clearest opportunity for governments to agree to launch negotiations on binding rules for autonomous weapons.
Guterres and Spoljaric said work carried out over the past three years had created a strong foundation for such talks and urged states not to allow that progress to be lost.
“We urge States to seize this moment,” they said.
The statement closed with a broader warning about the direction of military innovation, arguing that technologies making warfare easier risk undermining international efforts toward peace.
“Innovation must serve humanity – not endanger it,” the two leaders said.
Anthropic-Pentagon clash highlights AI weapons red lines
The UN and Red Cross warning comes months after Anthropic clashed with the Pentagon over how far the U.S. military should be allowed to use its Claude AI models.
The dispute emerged in February, when Anthropic refused to remove safeguards preventing Claude from being used to independently select and engage targets after the Pentagon sought access to the model for “any lawful use.”
The company argued that current frontier AI models are not reliable enough for fully autonomous weapons and could create serious risks for civilians and military personnel.
The standoff escalated in February when Defense Secretary Pete Hegseth moved to designate Anthropic a supply-chain risk after negotiations broke down.
Anthropic later challenged the move in court, turning the dispute into a wider debate over whether AI companies should retain the power to impose limits on how their technology is used by the military.
AI war simulations repeatedly ended in nuclear escalation
War simulations have raised similar concerns, with AI systems often responding to strategic pressure by escalating rather than backing down.
A 2026 King’s College London study tested GPT-5.2, Claude Sonnet 4 and Gemini 3 Flash across 21 simulated nuclear confrontations, with the models acting as rival leaders under strategic pressure.
Mutual nuclear threats or warnings appeared in nearly all of the simulations, while tactical nuclear weapons were used in 95% of the games, according to the study.
Researchers also found that none of the models chose outright surrender or withdrawal, even when facing intense pressure, while threats often triggered further escalation rather than compliance.
While there is still no clear indication of what level of firepower future autonomous weapons may ultimately control, the findings suggest AI systems can treat highly destructive capabilities, nuclear weapons in this case, as strategic tools rather than a moral red line, reinforcing concerns over giving autonomous systems greater influence in high-stakes military decision-making.
