Calls to put safety first in artificial intelligence development dominated an industry summit on Wednesday, with speakers raising concerns that ranged from AI worms spreading through household appliances to possible major security incidents in the future. The conference was held in Dublin.
Worries about AI guardrails have grown since models developed by OpenAI and Anthropic were involved in security incidents during testing. That has added pressure on lawmakers to regulate the technology.
Hassabis warns AI capabilities are advancing rapidly
Google DeepMind chair Demis Hassabis told the conference that AI capabilities were growing fast. He described the recent incidents as “early warning shots” fired by systems that “aren’t that powerful yet”.
Hassabis said the incidents were not especially serious at present, but could become so over the next few years if the industry, individual companies and governments do not do more to secure these systems. He added that new AI models should likely be tested first in offline “sandboxes”, since models from other companies had escaped the closed environments used to assess their capabilities.
Microsoft’s Brad Smith calls for urgent new safety laws
Microsoft president Brad Smith called for regulation, saying industry and governments must “act with urgency” to create “real and effective safety standards” that contain the risks linked to artificial intelligence. He contrasted the current situation with everyday products, noting that dairy goods and kitchen ovens face more safety rules than AI does.
Smith said the power of AI had advanced faster than the safety measures meant to govern it, and that new AI safety laws and rules would be needed. He proposed that AI agents be fitted with “safety brakes” that users could activate if an incident occurred, and that AI providers face mandatory incident-reporting obligations.
Anthropic report raises fears of self-copying AI worms
Smith pointed to a recent Anthropic report suggesting that small AI models may be only months away from behaving like computer worms, copying themselves from one internet-connected device to the next. Anthropic said that as tech companies build more powerful AI directly into phones, laptops, cars and household devices, such AI might eventually move autonomously from device to device.
Speaking at the Dublin conference, Smith voiced his alarm directly, saying, “I don’t want AI worms in my refrigerator.” Anthropic said a self-replicating AI worm would be difficult to stamp out because there would be no central switch to turn off.
EU and OpenAI weigh testing against innovation
The European Union’s tech chief, Henna Virkkunen, told the conference that AI firms needed to spend more time testing their systems in order to keep control over them. She told reporters that competition was so intense that companies did not appear to be investing enough time and resources in testing their models.
Virkkunen said the EU had sent information requests to all major AI companies asking what safety and security measures they have in place. The European Commission gained new regulatory powers in August under the AI Act, which the bloc describes as the world’s most far-reaching rules to curb potentially dangerous uses of the technology.
OpenAI CFO Sarah Friar warned that a “balance” was needed so that safety rules did not hold back innovation. She said the company wanted to be open and ensure the right people could review models, but that such reviews must not slow it so much that it fell behind those using the technology for malicious purposes.
