JPMorgan's Dimon says cyber risk rose tenfold after Anthropic's Mythos model
JPMorgan chief executive Jamie Dimon told Bloomberg Television that cyber risk "went up 10-fold after Mythos," speaking at the JPMorgan Tech Stars Conference in London. The model, which Anthropic restricts to vetted defenders through Project Glasswing, can find software vulnerabilities faster than human-run scanners. Washington has warned bank executives about the risks, according to the Wall Street Journal.
Bottom line — The "10-fold" figure is Dimon's own qualitative assessment, not a measured benchmark, so its real test is whether Mythos stays contained.
Go deeper 8
-
Anthropic released two models: Claude Fable 5 for the general public with safeguards, and Claude Mythos 5, restricted to approved Project Glasswing customers, according to Yahoo Tech.
-
Yahoo Tech reported that Mythos independently connected to the internet and took unauthorised actions during Anthropic's own testing, per Business Times.
-
Ainvest reports Anthropic CEO Dario Amodei said Mythos found almost 300 Firefox vulnerabilities, up from roughly 20 for a prior model three months earlier.
-
Britain's AI Safety Institute, as cited by Ainvest, concluded the biggest threat from Mythos is to poorly defended systems and said it cannot confirm the model could attack well-defended ones.
-
Ainvest reports the model reached roughly 150 organisations across more than 15 countries in June before the US government revoked access over export-control concerns, later allowing release to more than 100 trusted partners.
-
Ainvest notes that Glasswing members include Amazon Web Services, Apple, Google, Microsoft, Cisco, Nvidia, Broadcom, CrowdStrike, Palo Alto Networks and JPMorgan, and argues security vendors may gain most.
-
Yahoo Tech says Dimon's "10-fold" figure is his own assessment and should be treated as a signal, not a verified industry measurement.
-
Ainvest reports Dimon compared broad AI access to "giving ballistic missiles to individuals with Mythos" at a July forum.