Existing footprint and Singapore connection

• Anthropic is headquartered in San Francisco and already has an international presence, including European and Asian hubs such as Dublin, London, Zurich, and Tokyo; the Singapore hiring would extend that footprint into Southeast Asia.
• Singapore’s state investor GIC is a significant Anthropic backer. It led Anthropic’s $30 billion Series G with Coatue at a $380 billion post-money valuation, after having invested in the prior $13 billion Series F.
• The company subsequently raised $65 billion in Series H, at a $965 billion post-money valuation; GIC was among the round’s co-leads.

Mythos Preview and Project Glasswing

• Anthropic did not make Claude Mythos Preview generally public because it said the model could outperform humans in some cyber-security and hacking tasks, notably identifying software vulnerabilities. The core risk is dual use: a capability that can help defenders discover flaws could also let attackers find and exploit them at greater speed and scale.
• Project Glasswing was the controlled-access response: selected technology and critical-infrastructure organizations could use Mythos to scan important software for vulnerabilities rather than receive a broadly deployable model. The approach seeks to direct the capability toward remediation, although restricted distribution does not eliminate misuse, leakage, or concentration-of-capability risks.
• U.S. officials were sufficiently concerned about the possible cyber implications that the White House was reported to be preparing access for key federal agencies.

Pentagon policy dispute

• The dispute was not simply that Anthropic declined defence work. The reported conflict was over the US Defence Department seeking unconditional use of Anthropic technology, while Anthropic maintained policy restrictions—particularly around autonomous weapons and mass surveillance.
• The administration and Pentagon treated those restrictions as unacceptable in a national-security context: on February 27, 2026, President Trump directed federal agencies to stop using Anthropic technology and Defence Secretary Pete Hegseth said he would designate Anthropic a supply-chain risk to national security.
• The policy question is consequential: Anthropic’s position prioritizes retained safety constraints on high-risk military use, while critics in government argue that a contractor should not be able to limit operational use when national-security agencies judge the technology necessary. The available evidence supports a genuine policy conflict, not a purely technical procurement disagreement.

Existing footprint and Singapore connection

• Anthropic is headquartered in San Francisco and already has an international presence, including European and Asian hubs such as Dublin, London, Zurich, and Tokyo; the Singapore hiring would extend that footprint into Southeast Asia.
• Singapore’s state investor GIC is a significant Anthropic backer. It led Anthropic’s $30 billion Series G with Coatue at a $380 billion post-money valuation, after having invested in the prior $13 billion Series F.
• The company subsequently raised $65 billion in Series H, at a $965 billion post-money valuation; GIC was among the round’s co-leads.

Mythos Preview and Project Glasswing

• Anthropic did not make Claude Mythos Preview generally public because it said the model could outperform humans in some cyber-security and hacking tasks, notably identifying software vulnerabilities. The core risk is dual use: a capability that can help defenders discover flaws could also let attackers find and exploit them at greater speed and scale.
• Project Glasswing was the controlled-access response: selected technology and critical-infrastructure organizations could use Mythos to scan important software for vulnerabilities rather than receive a broadly deployable model. The approach seeks to direct the capability toward remediation, although restricted distribution does not eliminate misuse, leakage, or concentration-of-capability risks.
• U.S. officials were sufficiently concerned about the possible cyber implications that the White House was reported to be preparing access for key federal agencies.

Pentagon policy dispute

• The dispute was not simply that Anthropic declined defence work. The reported conflict was over the US Defence Department seeking unconditional use of Anthropic technology, while Anthropic maintained policy restrictions—particularly around autonomous weapons and mass surveillance.
• The administration and Pentagon treated those restrictions as unacceptable in a national-security context: on February 27, 2026, President Trump directed federal agencies to stop using Anthropic technology and Defence Secretary Pete Hegseth said he would designate Anthropic a supply-chain risk to national security.
• The policy question is consequential: Anthropic’s position prioritizes retained safety constraints on high-risk military use, while critics in government argue that a contractor should not be able to limit operational use when national-security agencies judge the technology necessary. The available evidence supports a genuine policy conflict, not a purely technical procurement disagreement.