As India gets Claude Mythos, attention turns to AI economics

Bengaluru: India's inclusion in Anthropic's Project Glasswing initiative marks a significant step for the country's cybersecurity and AI ecosystem. Access to Claude Mythos Preview, an advanced AI model designed for cybersecurity applications, is expected to strengthen vulnerability detection and threat response capabilities while raising broader questions about the economics of deploying agentic AI at scale.Experts say the technology represents a significant leap in cybersecurity capabilities. Srinivas Padmanabhuni, CTO of AiEnsured, said Mythos Preview's ability to identify zero-day vulnerabilities faster and more comprehensively than conventional tools could fundamentally change how organisations approach security."Traditional vulnerability assessment is largely reactive. AI-powered detection shifts the posture from reactive to proactive, which in cybersecurity is the difference between containing a breach and preventing one," he said.Padmanabhuni added that Anthropic's decision to include India among the 15 countries participating in Project Glasswing signals growing confidence in the country's role in deploying frontier AI responsibly."It accelerates the case for India being part of global conversations on AI governance, not just as a market, but as a contributor," he said.Viral Shah, CEO of JuliaHub, said India's inclusion in the initiative could also create opportunities for deeper collaboration between the country's technology ecosystem and leading AI research labs."Given the depth of the venture capital ecosystem and the connections between Bengaluru and Silicon Valley, it would be very interesting to see bridges and real collaborations being built between the top technology firms in India and the top AI labs," Shah said.He added that India's long-term competitiveness in AI would ultimately depend on infrastructure. "Progress in AI boils down to energy, data-centre construction and access to silicon. "These are areas where the Indian government can help streamline investments through initiatives such as Make in India," Shah said.While the Mythos model's capabilities have drawn attention, CloudSEK founder Rahul Sasi believes the bigger debate about Mythos will ultimately center on economics rather than technology alone."The most expensive part of LLMs in the long run is not training alone. Training is expensive but largely a one-time cost. The recurring expense is inference—hosting the model and making it reason, call tools and retry until it reaches a useful outcome. This is why token cost matters," Sasi wrote in a LinkedIn post.According to him, AI model providers are likely to push the challenge of token optimisation downstream to OEMs, SaaS companies and cybersecurity vendors. While foundation model developers will continue improving model performance, application-layer companies will have to solve problems around context management, retrieval, caching and workflow orchestration. "The real question isn't whether agents can hack. It's how many tokens, failed attempts, tool calls and dollars it took to get there," Sasi said.He added that companies automating security information and event management (SIEM) and security operations centre (SOC) workflows may find that AI agents are not necessarily inexpensive substitutes for human analysts. "An agent does not answer once and stop. It keeps reasoning, calling tools, summarising context, retrying and burning tokens. In some cases, that may cost more than a human analyst who already has business context and environmental knowledge," Sasi said.