Anthropic CEO Dario Amodei warned that rapidly advancing AI systems could trigger a major cybersecurity crisis if governments and enterprises fail to address software vulnerabilities uncovered by the company’s latest model. Speaking at an Anthropic event alongside Jamie Dimon, Amodei said the company’s experimental AI model, Mythos, has already identified tens of thousands of security flaws across widely used software systems.

Anthropic previewed Mythos last month, revealing that the model had discovered vulnerabilities that had remained undetected for decades. According to Amodei, the scale of findings has expanded dramatically with each generation of Claude models. Earlier systems identified dozens of vulnerabilities in individual applications such as Firefox, while Mythos uncovered nearly 300 in that browser alone. Across all software reviewed, the total now reaches into the tens of thousands.

The company has restricted access to Mythos because of concerns that hostile actors or cybercriminals could exploit the information. Many of the vulnerabilities remain undisclosed because patches are not yet available. Amodei warned that adversarial nations and malicious groups could use similarly advanced AI systems to identify weaknesses at scale once comparable models become widely accessible.

He also suggested that Chinese AI developers are only 6 to 12 months behind leading US systems, creating what he described as a limited window for organizations to strengthen their defenses. The concern is that AI-powered vulnerability discovery could accelerate ransomware attacks, data breaches, and other cyber threats targeting critical institutions including banks, hospitals, and schools.

Alongside the cybersecurity warning, Anthropic unveiled an expanded financial services platform that includes ten AI agents designed for investment banking and operational workflows, as well as broader integration of Claude across Microsoft Office applications. The announcements reinforced Anthropic’s growing push into enterprise AI markets.

A Growing Race Between AI And Cyber Defense

The warning highlights a growing tension within the AI industry. The same systems that can improve software development and automate security analysis can also expose weaknesses faster than organizations can fix them. AI-driven vulnerability discovery could significantly increase pressure on cybersecurity teams already struggling with patch management and infrastructure complexity.

For enterprises, the challenge is becoming less about whether vulnerabilities exist and more about how quickly they can be identified and remediated before attackers exploit them. Financial institutions, healthcare providers, and public infrastructure operators are particularly exposed because of their reliance on older software systems and interconnected networks.

Amodei’s comments also reflect broader concerns that AI capabilities are advancing faster than regulatory frameworks and defensive measures. While AI models can improve threat detection and incident response, they also lower the technical barriers for identifying exploitable flaws at scale.

Enterprise AI Expansion Meets Regulatory Pressure

Despite the risks, Anthropic positioned AI as a technology that could ultimately improve cybersecurity if managed responsibly. Amodei compared future AI oversight to safety regulation in the automotive industry, arguing for guardrails that allow innovation while limiting the most dangerous outcomes.

The company’s appearance alongside JPMorgan Chase underscored Anthropic’s increasing influence in enterprise AI. That push is extending beyond software models into deployment and consulting infrastructure. Anthropic is also partnering with Blackstone and Goldman Sachs on a planned $1.5 billion AI services venture aimed at helping portfolio companies integrate Claude into their operations, especially mid-sized firms without large in-house AI teams.

ElevenLabs has surpassed $500 million in annual recurring revenue and announced new investors as part of an expanded Series D funding round. The latest backers include BlackRock, Nvidia through its NVentures arm, and Santander, alongside creative figures such as Jamie Foxx and Eva Longoria. The announcement follows strong revenue growth, with the company reporting $350 million ARR at the end of 2025 and adding $150 million more in the first four months of 2026.

The growth is tied to increasing enterprise adoption of AI-powered voice agents across functions such as customer support, sales, hiring, and marketing. ElevenLabs said large organizations are deploying its tools to enable natural, human-like interactions at scale, often across multiple languages and channels. Institutional investors including Wellington Management and Schroders joined the round, signaling confidence in conversational AI as a core business infrastructure.

Several enterprise customers are also participating as investors. Companies such as Salesforce, Deutsche Telekom, and KPN are already using ElevenLabs’ platform for applications ranging from advertising and product demos to real-time translation and AI-driven customer support. Deutsche Telekom, for example, has deployed voice agents within its network to assist users during live calls, highlighting how telecom providers are integrating AI deeper into core services.

The company is also expanding its reach beyond enterprise and into the creative economy. More than 30 high-profile figures, including filmmaker Hwang Dong-hyuk and actor Matthew McConaughey, are backing the platform. ElevenLabs said these partnerships reflect growing interest among creators in tools that allow them to scale and localize their voices for new audiences and formats.

Enterprise Demand Fuels Voice AI Expansion

The rapid rise in revenue underscores how voice-based AI is becoming a key interface for businesses. Unlike text-based systems, voice interactions require low latency and high realism, making them technically demanding but valuable for customer-facing roles. ElevenLabs’ focus on generating natural-sounding speech positions it within a segment where quality directly affects user trust and engagement.

For enterprises, the appeal lies in automation without sacrificing personalization. AI voice agents can handle high volumes of interactions while maintaining consistent tone and language, which is particularly useful for global companies. The company’s expansion into multilingual capabilities and real-time applications suggests that voice is evolving into a primary communication layer across digital services.

The involvement of financial institutions and strategic investors also indicates that conversational AI is increasingly viewed as long-term infrastructure rather than an experimental technology. This could accelerate adoption across sectors such as banking, telecom, and retail.

From Voice Models to Full-Stack AI Communication

ElevenLabs began by developing high-quality AI voice models, but it is now positioning itself as a broader communication platform. The company plans to integrate audio with image and video generation tools, allowing businesses and creators to produce complete marketing and media assets within a single system.

On the enterprise side, the roadmap includes expanding AI agents beyond voice into chat, email, and other channels. This reflects a shift toward unified AI systems that manage customer interactions across touchpoints. The company is also investing in international expansion, building local teams to tailor deployments to regional markets.

Alongside its funding round, ElevenLabs completed a $100 million tender offer, its second within a year, providing liquidity to employees as the company scales. It currently has more than 500 employees across 50 countries.

Meta has introduced a new set of updates aimed at improving how it identifies and protects younger users across its platforms, including Instagram and Facebook. The company is expanding its use of artificial intelligence to detect underage accounts and enforce age-based protections, while also rolling out Teen Account safeguards to more regions. The changes come as regulatory scrutiny and public concern over youth safety online continue to increase globally.

A key part of the update is the use of more advanced AI systems to identify users under 13, who are not permitted on Meta’s platforms. The company said it is now using visual analysis alongside text-based signals to assess whether an account belongs to a minor. This includes scanning posts, captions, and images for contextual clues such as school references or birthday celebrations. If an account is flagged, it may be deactivated unless the user verifies their age through official checks.

Meta emphasized that the visual analysis does not rely on facial recognition. Instead, it evaluates general characteristics such as physical features and contextual elements in photos and videos to estimate age. The company is also deploying AI tools to assist with reviewing reports of underage users, aiming to improve both speed and accuracy compared with human moderation alone.

Alongside enforcement, Meta is expanding its system that automatically places suspected teens into restricted Teen Accounts. These accounts include built-in protections such as limits on who can contact users and what content they can see. After initial rollouts in the United States, Canada, the United Kingdom, and Australia, the company is extending these measures to 27 European Union countries and Brazil on Instagram. Facebook will also begin using the system in the US, with further expansion planned in the UK and EU.

Tighter Controls and Broader Coverage

The updates reflect a broader shift toward proactive age verification as platforms face increased pressure to protect younger users. By combining AI detection with automatic safeguards, Meta is reducing reliance on self-reported age, which has historically been unreliable.

For teens, this means more accounts will be placed into restricted environments by default, even if users attempt to bypass safeguards. Parents may also see more prompts and tools designed to encourage age transparency. For the wider industry, the rollout signals a move toward automated enforcement at scale, which competitors may need to match as expectations rise.

Meta’s call for app store-level age verification also points to a potential structural change in how age assurance is handled. A centralized system could simplify compliance and create more consistent protections across apps, though it would require cooperation from platform operators.

Policy Pressure and Platform Evolution

Meta’s latest measures build on years of investment in youth safety features, including Teen Accounts across Instagram, Facebook, and Messenger. These systems automatically apply stricter privacy settings and content limitations for users under 18.

The company has increasingly turned to AI to address moderation challenges, including detecting harmful content and identifying policy violations. Age detection remains one of the most complex problems, as users can misrepresent their identity and behavior varies widely across regions.

Similar approaches are emerging across the industry. OpenAI introduced age prediction tools within ChatGPT to adjust safety settings for younger users, while allowing adults to verify their age to access fewer restrictions. The parallel efforts highlight a growing consensus among major platforms that automated age assurance will play a central role in meeting regulatory demands and improving online safety for teens.