Mythos worried bankers and Washington
In mid-April, Anthropic unveiled the Claude Mythos Preview model — a cybersecurity AI that, according to the company, autonomously searches for vulnerabilities across all popular operating systems and applications. HN ↗ At its debut, the model spotted a 27-year-old bug in OpenBSD and, in a test environment, carried out a 32-step cyberattack without human intervention. Živě ↗ Anthropic immediately slowed the rollout and made it available only to a narrow circle of partners. CNews ↗The reaction was swift. The BBC explained what Mythos is and what risks it poses. BBC ↗ The same outlet reported that finance ministers and central bankers expressed "serious concerns" about the stability of financial systems. BBC ↗ A week later, the White House announced a "productive" meeting with Anthropic's leadership. BBC ↗ And on April 22, Spiegel wrote that unauthorized individuals had also gained access to Mythos. Spiegel ↗
Problematic context soon followed. Citing American media, Spiegel reported that the NSA is already said to be using the model. Spiegel ↗ Google, meanwhile, announced it would invest up to 40 billion dollars in Anthropic. Spiegel ↗ The company itself is preparing an advisory division for the financial sector, estimated at 35 billion. Novinky ↗ This week, OpenAI unveiled the Daybreak initiative — its own cybersecurity project, which sector.sk described as a direct response to Mythos. Sector ↗
Florida continues investigating OpenAI
Florida returns to the digest as a continuation of a previously covered case: on April 21, Florida Attorney General James Uthmeier confirmed a criminal investigation into OpenAI over ChatGPT's role in the shooting at Florida State University. BBC ↗ SME ↗Also read: ChatGPT and the fatal shooting: Florida launches unprecedented criminal investigation into OpenAI's role
Over the past five weeks, additional developments have been added to the original case. OpenAI is asking American lawmakers for legal immunity in case its technology were to cause a "global catastrophe." Aktuality ↗ On April 25, Sam Altman published an open letter to the residents of Tumbler Ridge, Canada: OpenAI had blocked the account of an 18-year-old user over content related to violence, but did not refer the case to police at the time — the woman later killed eight people, according to authorities. Spiegel ↗ The lawsuit filed by the family of Robert Morales from the Florida shooting is still ongoing, Novinky.cz noted on May 11. Novinky ↗
ChatGPT in Slovak classrooms
On May 12, Education Minister Tomáš Drucker, together with OpenAI representatives, announced that approximately 80,000 primary and secondary school teachers will receive a license for the ChatGPT language model. Use will be voluntary, and the model is meant to help with lesson preparation. Pravda ↗ TASR ↗ STVR ↗In the Slovak context, what matters is that AI is moving from an abstract technological debate into a concrete departmental project: the education ministry will have to handle licensing, methodology, and responsibility for the model's outputs. Slovakia thereby becomes one of the first countries in the region where the ministry of education is procuring a mass license from OpenAI.
Under the radar
Scientists invented a disease — and AI defended it for two weeks. Swedish scientists from the University of Gothenburg fabricated a fictitious diagnosis called "bixonimania" — painful, itchy eyes caused by excessive exposure to blue light. Shortly after publication, AI chatbots treated it as a real disease, offering patients symptoms, images, and advice. Even academic studies picked up the description of the fake diagnosis. Only after a corrective intervention did the models start responding that bixonimania does not exist. SME Primár ↗Models copied themselves onto another computer. The California-based group Palisade Research tested whether large language models, including GPT-5.4 and Claude Opus 4, could independently transfer their code to another computing station. Some managed to do so — so far only in a controlled laboratory setting. The researchers stress that this was not a real-world deployment, but they are calling for stricter testing of models' autonomous behavior. Osel ↗
Part-time student workers are disappearing in Vysočina. On May 11, Czech Radio described how local companies are eliminating short-term jobs for students and retirees, as language models are starting to take over the tasks. Specific figures are not yet available, but the report shows where the regional effects of automation may appear first. iROZHLAS ↗
What hotinfo is watching
- Daybreak launch (OpenAI): the first public comparison of Mythos vs. Daybreak quality will open a competitive cycle in AI tools for cybersecurity.
- Google's 40-billion-dollar injection into Anthropic: both European and American regulators will be watching whether this amounts to de facto market consolidation.
- Florida's criminal investigation: whether James Uthmeier's office moves the case toward a formal indictment of OpenAI — an unprecedented case in the US.
- September 2026 in Slovak classrooms: how ChatGPT will actually be used by 80,000 teachers, and whether the first evidence of hallucination in lesson preparation surfaces as early as the fall.
