Anthropic has released the upgraded Claude Opus 4.8 model while continuing to restrict public access to its powerful Mythos AI due to ongoing cybersecurity and safety testing requirements.

Key Points

• Anthropic launched Claude Opus 4.8, which offers improved benchmark performance and reduced hallucination rates compared to the previous 4.7 version.
• The company is currently limiting access to its advanced Mythos Preview model to select partners under the Project Glasswing initiative.
• Mythos demonstrated its utility by identifying over 200 security vulnerabilities for the latest version of the Mozilla Firefox browser.
• New features on Claude.ai and Claude Cowork allow users to adjust model "effort" settings to balance response quality against usage limits.
• Cybersecurity experts note that Mythos is roughly 30 times more expensive to operate than previous models, potentially limiting its use by smaller threat actors.

Why it Matters

The phased release of the Mythos model highlights the growing tension between rapid AI innovation and the need for robust safeguards against potential cybersecurity exploits. As these powerful tools become more accessible, the industry faces a critical window to implement defenses before the technology is widely adopted by both developers and malicious actors.