Mirendil’s $200M Bet on Self-Improving AI; US and Iran Agree to Stand Down

The AI funding race shows no sign of cooling. Mirendil, a startup founded by two former Anthropic researchers, has raised $200 million at a $1 billion valuation—before shipping a product—to build AI systems that automate AI research itself ^[1]. Andreessen Horowitz and Kleiner Perkins co-led the seed round, with Nvidia also joining, making it one of the largest seed financings the sector has seen ^[1].

Founders Behnam Neyshabur and Harsh Mehta left Anthropic in December 2025 after less than a year there; Neyshabur previously spent more than five years at Alphabet co-leading reasoning research for Gemini ^[1]. Mirendil’s pitch is to commercialise the recursive self-improvement loop that frontier labs keep in-house: software that designs experiments, tunes models, evaluates results, and iterates on training ^[1]. The company frames the platform as “AI for AI for science,” suggesting a biology lab could build a drug-target model without its own machine-learning team ^[1].

The deal arrives just weeks after Anthropic itself leapfrogged OpenAI as the world’s most valuable AI startup, closing a $65 billion round at a $965 billion valuation ^[3]. Anthropic also recently restricted access to its most powerful Mythos and Fable models following Trump administration export controls, a move that has intensified debate over who controls frontier AI capabilities ^[1]. Mirendil’s founders argue that self-improving AI can be supervised rather than locked away, a position that directly challenges the cautious stance Anthropic has taken ^[1].

On the world stage, Washington and Tehran have agreed to halt strikes in and around the Strait of Hormuz, according to a US official cited by CBS News and Reuters ^[2][4]. The stand-down follows several days of tit-for-tat attacks that threatened a 14-point ceasefire signed on 17 June, which included an “immediate and permanent termination of military operations on all fronts” ^[2]. The latest flare-up began Thursday when an Iranian projectile hit a cargo ship; the US retaliated over the weekend, and Iran then struck at US bases in Kuwait and Bahrain, though the US said those attacks caused no casualties or damage ^[2].

Under the renewed understanding, vessels will be allowed to move through the Strait of Hormuz “freely” and technical talks on the memorandum will resume ^[2][4]. Iran has not publicly confirmed the agreement ^[2]. The Strait, a chokepoint for global oil and gas shipments, had been effectively closed by Tehran after US and Israeli attacks on Iran in late February ^[2].

The two stories share a common thread: both centre on who controls access to powerful, contested capabilities—whether AI research automation or a strategic waterway—and how quickly fragile agreements can fray under pressure.