Anthropic Curbs Rollout of AI Skeleton Key While Meta Regains Model Momentum
Meta launched its first major AI model after a significant overhaul, while Anthropic is keeping its own dangerously good model under wraps.
Sign up for smart news, insights, and analysis on the biggest financial stories of the day.
One is positioning itself as the world’s new front door to AI; the other is a lock-pick so effective its creators are hiding it.
Meta reasserted itself in the artificial intelligence race Wednesday with the launch of its first new model since CEO Mark Zuckerberg poured billions into a major overhaul to catch rivals Google, OpenAI and Anthropic. The release came a day after Anthropic said it would severely limit the rollout of an advanced new model designed to excel at identifying software vulnerabilities because it would be a gift to the world’s hackers.
Open to the World vs. Close to the Vest
Meta’s new Muse Spark, developed by a lavishly compensated unit Zuckerberg assembled under wunderkind Scale AI cofounder Alexandr Wang, now powers Meta’s AI app and website in the US, with Facebook, Instagram, Messenger, WhatsApp and Meta smart glasses to follow in the coming weeks. Similar to Google’s Gemini model, it was “purpose-built for Meta’s products,” enabling the company to leverage its mass social media user base to expedite adoption. Meta is deliberately restraining expectations, highlighting “competitive performance” against ChatGPT, Gemini and xAI’s Grok. Meanwhile, it plans to put $115 billion to $135 billion toward capital investments including AI, nearly double last year’s tally, signaling its determination to become a leader in the space. Meta soared 6.5% on Wednesday, beating the S&P 500’s 2.5% bounce on positive news about the Iran war.
The splashy launch contrasted with Anthropic’s unusual announcement Tuesday that its advanced model, Claude Mythos Preview, would be distributed only to a few dozen companies, including Amazon, Apple, Microsoft, Nvidia and JPMorganChase, as part of its cybersecurity-oriented Project Glasswing. That’s undoubtedly for the best:
- Claude Mythos Preview was designed to detect and patch critical software vulnerabilities, securing digital infrastructure from AI exploitation before it’s too late. Anthropic said it has already identified thousands of vulnerabilities in critical software infrastructure, including a 27-year-old bug in the operating system OpenBSD that had gone unnoticed.
- Anthropic’s Mike Krieger told a conference in San Francisco that the model was “explicitly not releasing to the public,” and that the company plans to work with cybersecurity engineers and experts to use the model as a means of “arming them ahead of time.”
Partner Perks: When reports first circulated late last month about Mythos, some cybersecurity came under pressure over fears Anthropic would become a competitor. But JPMorgan analysts said Wednesday that CrowdStrike and Palo Alto Networks, both part of Project Glasswing, could benefit because they’re working closely with Anthropic on the new model. The bank affirmed its overweight ratings on both, with 12-month price targets of $475 for CrowdStrike and $200 for Palo Alto Networks, implying double-digit upsides for each.
