The fear around Mythos is also self-serving AI hype. For India, safety might lie in more open source resources

The debate over AI model release policies has historical precedent. In 2019, OpenAI initially claimed that GPT-2 was too dangerous to release publicly, though this position was later deemed by many experts to be unjustified. This pattern of claiming frontier AI models are too dangerous for public release has now been adopted by Anthropic with Mythos.

The concept of "security through obscurity" versus open security has deep roots in cybersecurity discourse. Proprietary software has traditionally relied on keeping source code hidden, while Free and Open Source Software (FOSS) has operated on the principle that openness enables better security — famously captured in Linus's Law: "Given enough eyeballs, all bugs are shallow."

The emergence of LLMs capable of analyzing code and finding vulnerabilities represents a new frontier in cybersecurity. The National Technical Research Organisation (NTRO) in India handles technical intelligence, while CERT-In (Indian Computer Emergency Response Team) coordinates cybersecurity responses. The IT Act, 2000 and its subsequent amendments provide the legal framework for cybersecurity in India.

The current geopolitical context involves US-China technology competition, with both nations developing advanced AI capabilities. India's National AI Strategy and the IndiaAI Mission represent efforts to build domestic AI capacity. The debate over digital sovereignty — whether nations can maintain autonomous control over critical technology infrastructure — has become increasingly prominent in policy discussions.

Mythos is Anthropic's proprietary LLM deemed "too dangerous to release" because it discovers security vulnerabilities in software (Source: Article). Project Glasswing provides exclusive access to Mythos for the US government and select American software companies (Source: Article). Anthropic reportedly allocated $100 million for the Mythos vulnerability-finding project (Source: Article). Cybersecurity firm Aisle found that many of Mythos's publicized findings could be replicated using smaller, cheaper open-weights models (Source: Article). A paper by Hanzhi Liu et al demonstrated that coordinated open-source models (Kimi K2.5) found 10 zero-day vulnerabilities in Google Chrome (Source: Article). US Big Tech companies have compared advanced AI models to "digital nukes" to erect regulatory barriers against competition from Chinese and Indian companies (Source: Article). Most successful cyber-attacks exploit known-but-unpatched vulnerabilities, not zero-day vulnerabilities (Source: Article). The principle "given enough eyeballs, all bugs are shallow" (Linus's Law) represents the FOSS security philosophy that openness enables better security (Source: Article).

Analyze the implications of India's dependence on proprietary foreign AI models for critical infrastructure security with reference to digital sovereignty. (GS-III, 250 words) Critically examine the debate between "security through obscurity" (proprietary software) and open-source security models in the context of AI-powered cybersecurity. (GS-III, 250 words) Discuss whether the rhetoric of AI models being "too dangerous to release" serves protectionist purposes rather than genuine security concerns, with reference to Project Glasswing. (GS-II, 250 words) Evaluate the supply-chain risks associated with relying on proprietary LLMs optimized for single-country chip architecture for national cyber-defence. (GS-III, 250 words) "Given enough eyeballs, all bugs are shallow." Examine how this principle applies to AI-powered vulnerability discovery and its implications for India's cybersecurity strategy. (GS-III, 250 words) Discuss the role of open-source AI models in achieving Atmanirbhar Bharat in critical technology infrastructure, with reference to the Mythos debate. (GS-II, 250 words)