Portuguese 
Jacob Coxon Accused Companies of Racing Towards Self-Improving Systems as International Report Acknowledges Rapid Advances but States Current Models Still Don’t Pose Total Loss of Control Risk
The risk of artificial intelligence has returned to the forefront of debate after Jacob Coxon announced his departure from Anthropic and accused major companies in the industry of developing increasingly powerful systems without sufficient safety guarantees. The episode has also revived warnings made by Stephen Hawking before his death in 2018.
Coxon stated that he worked for three years on pre-training research at Anthropic and OpenAI. According to the researcher, both companies are racing towards the development of a superhuman AI capable of self-improvement.
Hawking was already concerned about machines surpassing humans back in 2017. That year, the British physicist expressed fears that a self-replicating AI could become a new form of life and completely surpass humanity.
-
Taiwan Prepares Additional $20 Billion in Business Investments in the U.S. Amid AI-Powered Global Chip Race
-
GlobalFoundries Develops Test Chip with New Memory: 40% Smaller and Up to 60% More Energy Efficient, Targets AI, Cars, and IoT for 2027
-
Europe Orders €387.8 Million AI Supercomputer, Aims to Boost Capacity Tenfold by 2027 in Finland
-
Historical Barão de Anhumas Farm to Become 400 MW AI Data Center in Campinas After Over 200 Years in Sugar, Coffee, Cattle, and Heritage Preservation
Despite the warnings, there is currently no scientific evidence that the models available today could cause human extinction. International assessments recognize significant advances in autonomy and cybersecurity, but classify scenarios of total loss of control as uncertain and future risks.
Researcher Leaves Anthropic and Issues Public Warning
Jacob Coxon announced his resignation on September 8, 2026, via a post on platform X. The researcher stated that he did not wish to continue participating in a competition between companies eager to build self-improving systems.
His departure and statements were reported by the Associated Press. According to the agency, Coxon accused Anthropic and OpenAI of prioritizing competition among companies and countries in their quest to develop the most advanced model possible.
In his statement, the researcher asserted that companies are “betting with our lives.” He also expressed belief that future systems could achieve capabilities beyond human levels, launch cyberattacks, and acquire resources in the real world.
These assertions represent Coxon’s personal projections. The possibility of AI threatening humanity by the end of the decade has not been confirmed as a guaranteed event nor established as a scientific prediction.
Anthropic, for its part, has stated that it works to prioritize safety when this goal conflicts with the speed of development. The company has built its public image around producing reliable models and researching methods aimed at reducing dangerous behaviors.
Risk of Artificial Intelligence Concerned Hawking
Stephen Hawking raised various warnings about the future of technology. In an interview published by Wired on November 28, 2017, the physicist asserted that the development of artificial intelligence should continue but must take real dangers into account.
Hawking expressed concern that technology could completely replace humans. He cited the possibility of someone developing an AI capable of replicating itself, creating a new life form that could perform better than humans.
This statement was made roughly three and a half months before the scientist’s death on March 14, 2018. However, Hawking did not set a timeline for the emergence of superhuman AI nor claimed that extinction was inevitable.
The physicist portrayed the scenario as a possibility contingent upon the future development of technology. His concern was primarily linked to systems that could evolve, replicate, and surpass the limitations imposed by human biological evolution.
Current Systems Still Have Significant Limitations
The 2026 International AI Safety Report, prepared by over a hundred experts and backed by representatives from more than 30 countries and international organizations, confirms that the capabilities of AI models have advanced rapidly.
The most significant progress has been observed in mathematics, programming, sciences, and autonomous operation. Current systems are also capable of identifying digital vulnerabilities, producing malicious code, and performing some tasks for extended periods without constant intervention.
However, the document emphasizes that these capabilities remain inconsistent. Models capable of solving complex problems still fail at seemingly simple tasks, make errors, fabricate information, and struggle to correct failures during prolonged activities.
Regarding a potential loss of control, the report clearly states that current systems lack the necessary capabilities to create such a scenario. At the same time, it acknowledges that models are advancing in areas related to risk, such as autonomy, gap identification, and adaptation during evaluations.
The future remains uncertain. Development may slow due to limitations in energy, data, or infrastructure, continue at the current pace, or accelerate if the systems themselves begin to contribute significantly to new artificial intelligence research.
Thus, the warnings from Hawking and Coxon describe future possibilities seriously considered by the scientific community. They do not indicate that a superhuman artificial intelligence currently exists or that the disappearance of humanity is imminent, but they reinforce the ongoing debate about safety, control, and responsibility in the development of more advanced models.
Source: LADBIBLE
