Concerns about powerful new AI tool
The unveiling of Anthropic’s new Claude Mythos AI model caused alarm across the global banking and cybersecurity sectors, but experts said the prohibitive cost meant it was too early to panic.
First revealed on 7 April 2026, Mythos was touted by Anthropic as having the potential to “reshape cybersecurity.”
More specifically, Anthropic claimed that during testing, Mythos found and exploited thousands of zero-day vulnerabilities across every major operating system and web browser tested.
This raised alarms around the rapid development of large-language AI models and the threat they could potentially pose to cybersecurity systems.
Banks in South Africa and around the world questioned whether models similar to Mythos could disrupt the stability of the global financial sector.
To allay these concerns, Anthropic said it will not make Mythos available to the general public, but only to selected partner organisations.
Through Project Glasswing, Anthropic and its partners will deploy Mythos to detect and fix potential vulnerabilities in critical cybersecurity systems before they can be exploited in cyber attacks.
“Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely,” Anthropic said.
“The fallout — for economies, public safety and national security — could be severe. Glasswing is an urgent attempt to put these capabilities to work for defensive purposes.”
Partners for Project Glasswing include industry titans such as Microsoft, Apple, Google, Nvidia, and Amazon Web Services.
Major cybersecurity companies, including Cisco, Palo Alto Networks, and CrowdStrike, were also announced as being involved.
Anthropic said they had engaged in discussions with officials from the United States government on the potential offensive and defensive capabilities of Mythos.
The company previously denied the US Department of Defense access to its Claude model for use in mass surveillance of American citizens, as well as for developing fully autonomous weapons.
Too early for panic
While the announcement of Mythos’ capabilities sent shockwaves through the global banking sector, experts said there was no cause for immediate alarm.
In an interview with CapeTalk, Arthur Goldstuck, an industry analyst and founder of World Wide Worx, discussed whether the concerns raised held much weight.
“They are well-founded, but the level of urgency perhaps isn’t as great as it’s been made out to be,” Goldstuck said.
“We have a little bit of leeway because Anthropic themselves have realised how powerful this tool is, so they are not making it publicly available.”
In the meantime, cybersecurity experts such as Orange Cyberdefense South Africa managing director Dominic White urged companies to remain calm.
White said that until more information on Mythos and its capabilities becomes widely available, companies should not panic and hastily restructure their cybersecurity systems and strategies.
He also raised the issue of pricing, using token pricing from Anthropic’s publicly available Opus 4.6 model as a point of comparison.
“We don’t have costs for Mythos since it isn’t released, but 100M tokens (assuming an 80/20 input/output token split) would cost around $900 using Claude Opus 4.6,” White said.
“Sure, this is pocket change for large corporations or nation states, but assuming Mythos would be priced higher, $1k-ish per single attack doesn’t scale to ‘burn the whole Internet’ easily.”
White explained that the model’s potential unreliability could further drive up costs for attackers, as multiple attempts may be required before a zero-day is successfully discovered and exploited.
Seralys security researcher Philippe Caturegli also called for a “wait and see” approach, saying Anthropic’s marketing of Project Glasswing may have overstated Mythos’ true capabilities.
Caturegli pointed out that all testing of the model had been conducted internally by Anthropic and said more independent testing was needed to verify their claims.
“The technology might well be groundbreaking, but let’s not jump to conclusions and wait for independent validation before rewriting history,” Caturegli said.
Loading ...