Powerful artificial intelligence tools capable of finding and exploiting software weaknesses create the possibility of systemic cyber risk, says Arch Capital’s chief executive.

In an earnings call with analysts, Nicolas Papadopoulo singled out Mythos, an advanced AI tool built Anthropic for cybersecurity and coding tasks, as an example of AI’s “real, current threat”.

In testing Mythos uncovered more than 2,000 previously unknown software vulnerabilities in the space of seven weeks, including one that had existed in OpenBSD software for 27 years.

Recognising the tool’s potential as a weapon for cybercriminals, Anthropic has restricted the release of Mythos to a few major tech and cybersecurity firms.

Arch, a Bermudian-based global insurer and reinsurer, is a provider of cyber insurance and reinsurance.

“In terms of recent AI, Anthropic’s Mythos, we see it as a real, current threat,” Mr Papadopoulo said.

“We are seeing an arms race between attacker and defender. And certainly, Mythos is accelerating that trend. Mythos can help the attacker, but the defender can also reinforce its defences using the same model.”

He added that AI could make cyberattacks faster and bigger.

“Because the scale would be larger, I would think we see it as an increased systemic risk. So, we are taking a very careful approach to that in our RDS [realistic disaster scenarios].”

Mr Papdopoulo also detailed how Arch had benefited from using AI, and said the company had been investing in this technology for about a decade.

“We’ve deployed a bunch of AI and machine-learning models, but it’s changing really fast,” the CEO said.

He said the industry’s challenge was “to really show results while at the same time, working on our data strategy and integration of our system to really support AI at scale”.

Much investment was needed, while “at the same time, you’re trying to create productivity and the insight for your underwriters to be able to compete”, Mr Papadopoulo added.

With technological advances happening so rapidly, it was difficult for industry leaders to envisage what would be possible in three years, he said, adding: “When you look at the Anthropic models, they open huge opportunities to do certain things, but what’s next?”

Mr Papadopoulo highlighted how AI had helped to complete the data and system migration of acquired businesses from Allianz to Arch’s systems within 18 months.

AI had accelerated the building of the new platform by writing code and testing the new software, he said.

During the call, Mr Papadopoulo also commented on the softening property catastrophe reinsurance market.

“We have huge headwinds because of the double-digit rent decrease in rates,” he said. Arch monitored the property cat market through a lens of 50 separate zones, he added.

“Go back two years and they were all green,” Mr Papadopoulo said. “Now we have a bunch of them that are still green, including Florida, but some are yellow and some of them have turned red.”

On the casualty side, Mr Papadopoulo said Arch was still “optimistic”, noting several years of adverse development — meaning actual claims and payouts on policies turning out to be higher than initially estimated, forcing insurers to increase their loss reserves.

“We think the pain is not gone through yet,” he said. The years of adverse development “should, in our view, continue to sustain price increases above trend”, he added.

François Morin, Arch’s chief financial officer, was asked by an analyst about the company’s investment exposure to private credit, which he described as “limited”. Arch’s investment team had sought higher-quality loans with low loan-to-value ratios and good collateral, he added.

“It’s something we’re watching like everybody else,” Mr Morin said. “But at this point, there’s no red flags, nothing that is rising to a level where we have to take action.”