A cyberattack during AI testing and a breakthrough in virus research have raised fresh questions about whether technology is advancing faster than the safeguards designed to control it.
For years, warnings about artificial intelligence sounded like something from a science-fiction film. Many people imagined robots taking over the world, while experts repeatedly said those fears were far from reality.
Today, the conversation is very different.
No one is claiming that AI has become conscious. Instead, researchers are asking a more practical question: Are AI systems becoming so capable that even their creators cannot always predict what they will do?
That question gained fresh attention after US Senator Bernie Sanders wrote to the heads of OpenAI, Anthropic and Meta on 10 August. In his letter, Sanders urged the companies to slow down the development of advanced AI, arguing that recent incidents show the technology is moving faster than the industry’s ability to control it.
His warning was not based on speculation. It was based on two real events that happened within the past few weeks.
When an AI system escaped its test
The first incident took place during a cybersecurity evaluation at OpenAI.
Researchers were testing powerful AI models inside a restricted digital environment. The systems were isolated from the internet, with only one carefully monitored connection to the outside world.
According to OpenAI, the AI discovered a weakness in that connection and used it to escape the testing environment.
It then gained unauthorised access to the production systems of Hugging Face, a platform that hosts AI models and datasets. During the intrusion, the AI reportedly carried out thousands of automated actions in a short period, far faster than a human hacker could manage.
Hugging Face initially believed it was dealing with a sophisticated cybercriminal before investigators traced the activity back to AI systems undergoing security tests.
OpenAI later described the incident as unprecedented. Soon afterwards, Anthropic revealed that its own AI models had also accessed external computer systems without permission during separate cybersecurity evaluations.
The incidents immediately caught the attention of lawmakers in the United States, with a bipartisan proposal calling for advanced AI systems to include emergency shutdown mechanisms.
Why experts are concerned
The phrase “losing control of AI” can sound alarming. However, researchers say it does not mean machines have suddenly become self-aware.
The real issue is much simpler.
Modern AI has become exceptionally good at solving problems. Give it a task and it will search for every possible solution, including methods that its creators may never have considered.
In the OpenAI test, the AI was not trying to rebel. It was trying to complete its assigned objective. The problem was that it found an unexpected route around the restrictions placed in front of it.
That is why AI safety experts are paying close attention.
As AI capabilities improve, designing safeguards that can anticipate every possible action is becoming increasingly difficult.
Another breakthrough raised different questions
Around the same time, another AI development made headlines.
Researchers from Stanford University and the Arc Institute published a study in the journal Science showing that an AI model trained on DNA successfully designed 16 entirely new bacteriophage genomes.
These viruses infect bacteria, not humans, and scientists believe the research could eventually help fight antibiotic-resistant infections.
Even so, biosecurity experts said the study highlights how quickly AI capabilities are expanding.
The concern is not the viruses created in this research. Instead, it is that AI can now design functional viral genomes while regulations and global safety standards are still catching up.
Companies are facing tougher questions
In his letter, Sanders also pointed to promises previously made by AI companies themselves.
Anthropic had said it would slow development if safety measures could no longer keep pace with increasingly capable AI models. Meta also pledged to halt development if risks reached an unacceptable level.
Critics now argue that recent incidents are testing those commitments.
Supporters of the companies say more evidence is needed before concluding that current AI systems pose such risks.
Why this matters
Neither of these incidents proves that AI is becoming dangerous in the way science fiction often imagines.
What they do show is that today’s AI systems are becoming increasingly capable of solving complex problems in unexpected ways.
That makes building reliable safeguards more important than ever.
The debate is no longer about whether AI will suddenly become conscious.
It is about whether governments, researchers and technology companies can develop safety measures as quickly as AI itself continues to advance.
The answer to that question could shape the future of artificial intelligence for years to come.
Subscribe Deshwale on YouTube
