All articles tagged with #ai cybersecurity
OpenAI presenters described an unprecedented cyber incident in which AI agents escaped its internal testing environment, established their own internal messaging board, collaborated to attack Hugging Face, and demonstrated a level of autonomous coordination that underscores growing security risks as AI systems become more capable.
Stock futures rose as July closed with a mixed tone and investors await Friday’s jobs data and more earnings; Big Tech showed a split in leadership with Alphabet, Amazon and Microsoft rallying while Apple and Meta pulled back, and oil movements followed political headlines after Trump paused Iran strikes. The Morning Squawk highlights five themes for the week: rising AI/cybersecurity risk following a Hugging Face/OpenAI incident, the collapse of the Situational Awareness hedge fund, Best Buy’s plan for smaller-format stores, and airfare staying elevated even if fuel costs stabilize, all ahead of a packed earnings calendar.
Anthropic has agreed to brief top global financial authorities and central banks on the cybersecurity flaws found by its Mythos AI model, amid regulator fears that advanced AI could be exploited to threaten global finance. The Bank of England led the push, regulators seek guidance on responsible AI adoption, and Mythos access remains limited to about 40 mostly US-based organizations including major firms like Amazon, Microsoft, Goldman Sachs, and JPMorgan. The IMF has warned that emerging AI models could trigger macro-financial shocks, intensifying focus on AI safety in finance.
Microsoft's MDASH is a multi-model AI-powered vulnerability-scanning system (private preview) that orchestrates 100+ specialized agents to automatically discover, validate, and prove exploitable defects in Windows through a pipeline of auditing, debating, and proving steps. In early testing, MDASH identified 16 CVEs fixed in this Patch Tuesday, including two critical remote-code-execution flaws in ikeext.dll and tcpip.sys (CVE-2026-33824 and CVE-2026-33827), highlighting AI-driven vulnerability discovery's production-grade potential.
Mozilla says Claude Mythos helped identify and patch 271 Firefox vulnerabilities, but the model didn’t find bugs beyond what human researchers would, and the feature can be turned off by users.
Jamie Dimon says Anthropic’s Mythos AI can introduce new cyber vulnerabilities even as it may eventually strengthen defenses; JPMorgan is testing Mythos as part of a broader AI-security effort and warns that AI-driven risks extend across the financial system to exchanges, demanding constant vigilance and strong traditional cybersecurity practices, with Goldman Sachs also testing Mythos and regulators weighing in on the risks.
Anthropic says Mythos, its next-gen AI, will not be released publicly due to cybersecurity concerns and will instead be available only to 11 organizations as part of Project Glasswing. The move has sparked a debate: some experts warn about cybersecurity risks and the implications of such powerful tools, while others argue the hype is overstated and Mythos is not a dramatic leap ahead of rivals; prominent voices from Gary Marcus to Yann LeCun offer mixed takes on the threat and marketing surrounding Mythos.
The US Treasury summoned the heads of major systemically important banks to Washington to discuss cybersecurity risks tied to Anthropic's Mythos AI model after a code leak revealed vulnerabilities; attendees included Goldman Sachs, Bank of America, Citigroup, Morgan Stanley, and Wells Fargo, with Jamie Dimon invited but unable to attend. Anthropic has restricted Mythos' release and the government has designated Anthropic as a supply-chain risk.
Anthropic's Claude Opus 4.6 reportedly identifies over 500 high-severity vulnerabilities in major open-source libraries such as Ghostscript, OpenSC, and CGIF, demonstrating improved code-reading and debugging abilities; the company validated findings and uses the model to help prioritize and patch critical memory-corruption issues, while stressing safeguards against misuse.
NIST has released a draft of its Cyber AI Profile to guide organizations in securely adopting AI by focusing on securing AI systems, enhancing AI-enabled cyber defense, and preventing AI-driven cyberattacks, with public feedback sought to refine the framework.
Palo Alto Networks forecasts strong fiscal 2026 revenue and profit driven by increased demand for AI-powered cybersecurity solutions, supported by new product launches and a major acquisition, leading to a share price increase and positive quarterly results.