The Pentagon is running a new set of experiments designed to fold artificial intelligence into the fabric of air combat, testing autonomous "wingmen" and decision-assist systems that officials say will augment pilots and extend U.S. reach. CBS News’ recent coverage showed sorties in which crewed fighters flew alongside unmanned platforms that navigated, communicated and, in limited scenarios, executed missions with minimal human intervention.

Defense officials told CBS that the aim is not to replace pilots but to change how missions are planned and executed. "The goal is to give commanders tools that compress the decision cycle and keep human judgment where it matters," an Air Force spokesperson said. Program managers point to initiatives such as Skyborg and earlier loyal-wingman prototypes as precursors: cheaper, expendable aircraft that can take on high-risk tasks and expand the U.S. tactical envelope.

Proponents emphasize tangible operational gains. Autonomous systems can sustain persistent surveillance, act as decoys, and execute rapid maneuvers faster than humans can. That speed matters against peer competitors with sophisticated air defenses. "Autonomy can reduce pilot workload and enable formations that would otherwise be impossible," a Pentagon official said in an interview with CBS. Advocates also highlight cost advantages: unmanned platforms may be cheaper to produce and replace, allowing commanders to accept greater risk against hardened targets.

But researchers and ethicists caution that promise carries real hazards. Autonomous decision-making introduces new failure modes, from software bugs to adversarial machine-learning attacks that could mislead sensors. Testing in controlled ranges does not, they say, recreate the chaos of contested battlefields where degraded communications and deliberate deception are routine. "We are compressing time for decisions without yet fully understanding who bears responsibility when machines act," an academic specialist in defense policy told CBS.

Legal and oversight questions have moved from abstract debate into the halls of Congress. Defense Department directives dating back to 2012 require human judgment in the use of lethal force, and officials say current tests adhere to those rules by keeping humans in supervisory roles. Still, lawmakers and watchdogs are probing whether software that selects or prioritizes targets could erode meaningful human control.

Technical challenges persist. Autonomous aircraft must be robust to cyberattacks, interpretable to human operators, and verifiable in performance—a tall order for opaque neural networks trained on limited datasets. Engineers acknowledge that rigorous assurance regimes, rigorous simulation, and extensive live testing are necessary before such systems are fielded at scale.

Internationally, the experiments are likely to accelerate debates about norms and arms control. Allies watching U.S. development will confront similar trade-offs, and adversaries may pursue their own autonomous strategies. The Pentagon’s approach, officials say, is to develop interoperable systems that can operate with partners while remaining subject to clear policy guardrails.

As these trials proceed across U.S. test ranges and labs, the debate over how to blend human judgment with artificial speed will intensify. The coming years will determine whether autonomy transforms the air force into a network of human-machine teams or amplifies risks that policymakers have only begun to reckon with.