Meta (NASDAQ: META) is coming after ChatGPT and Claude with two new artificial intelligence models, and Mark Zuckerberg wants customers to start opening their wallets.This company has developed a product line called Muse Wonder and Muse Wonder Full. These products are designed for people looking for AI assistants capable of managing everyday work and developers who require extra computing power.The more sophisticated features of this system require monthly payments, unlike other products where Meta gives away its AI technology for free. Once again, this is an effort to create something else apart from ad-based income sources.The regular Muse Wonder model is available through Meta’s Muse personal agent app, with subscriptions costing $20 or $100 per month. The app itself remains free, although paying customers get more usage and access to the company’s latest technology.Meta launched Muse recently, and it soon rose up in Apple’s (NASDAQ: AAPL) App Store rankings. The challenge now will be to get the interested customers to believe that it is worth spending money on.Meta wants its AI agents to stop giving up when things get complicatedMeta’s AI chief, Alexandr Wang, said Muse Wonder, previously known internally as Watermelon, was built to make personal and workplace assistants better at handling unexpected problems.Think about asking an assistant to book dinner, only for the restaurant to have no available tables. Instead of freezing or abandoning the request, Meta wants its agent to consider other options and continue working.“When you do real-world tasks, you’re going to get roadblocks and you’re going to get things that come up or various setbacks,” Alexandr said. “We specifically tried to train the model to be able to try other approaches instead of stalling and then checking back in with the user at any time.”But giving AI assistants greater independence creates another headache: privacy and security.Muse users have also complained through social media that the agent accessed their personal messages without consent. Meta denies those claims, though Alexandr noted that his team is conducting investigations into the incidents being reported as well as monitoring the software.“I think there are certainly instances users have authorized various communications or whatnot earlier in a conversation, and then may have forgotten about that,” Alexandr said.Meta is trying to make those permissions easier to understand and educate customers about using the application safely.The company also says Muse Wonder includes its strongest protections against malicious cyber activity yet. Those additional safeguards are currently limited to the paid model, meaning free users don’t receive the same offering.Zuckerberg’s subscription push faces a bigger AI security problemMeta is not just stopping at consumers. Developers are able to get hold of Muse Wonder using Meta’s Model API; however, the cost will be greater than regular subscription plans.Alexandr stated that this was basically a bigger AI model that was frontier-level, and therefore required far more computing power to run. However, he insisted that Meta’s developer prices will still be lower than those of their competition, including OpenAI and Anthropic.The second release, Muse Wonder Full, goes further. This more powerful version is reserved for developers subscribing to Muse Code, Meta’s software engineering assistant, which offers $50 and $200 monthly plans.That puts Meta into direct competition with companies already selling AI-powered coding tools and autonomous agents. The business opportunity is clear, but so are the risks.OpenAI, Anthropic, and other tech firms’ executives have been privately debating how things would turn out in case an artificial intelligence disaster occurred.The first scenario is one where there could be an AI-driven attack that disrupts either financial services, internet connectivity, electricity, or water supply.Such an incident could trigger public outrage and political demands for tougher oversight, putting executives, including Sam Altman and Dario Amodei, under pressure. Trump could also face criticism over his reluctance to impose stricter AI regulations.Some leading researchers believe a serious incident is inevitable. However, OpenAI says preparing for dangerous possibilities doesn’t mean the company expects them to happen.“OpenAI conducts preparedness exercises where teams discuss and work through a range of potential scenarios,” a company spokesperson said. “These scenarios are not treated as inevitable, but are meant to help us prepare for a variety of circumstances.”Scenario planning is nothing new. The Pentagon has been doing it for decades. The only difference is that some of those who design AI systems see these threats as becoming a reality someday.If you're reading this, you’re already ahead. Stay there with our newsletter.