Mythos

Anthropic

2026-08-18 08:13:25

SemiAnalysis podcast says Anthropic tightened Fable access under regulatory pressure

A SemiAnalysis podcast said Anthropic placed unusually strict classifiers on Fable, the public version of its flagship Mythos model, as regulatory pressure shaped how the company released the system. According to the discussion, users attempting certain actions are immediately downgraded to the Opus model, and that downgrade cannot be reversed within the conversation. SemiAnalysis founder Dylan Patel said the restrictions were so severe that even internal staff could not use the model for routine work. The podcast also said Mythos had completed training in February but was not released under the Fable name until May. In the same discussion, Jordan said his understanding was that Mythos 2 had also finished training but had not been launched, while internal teams were still iterating. The speakers argued that Anthropic, which had long called for regulation, is now seeing that regulatory framework slow the pace of its own public model releases. They drew a parallel with OpenAI, saying its next-generation public release is also being delayed even as internal training continues, potentially widening the gap between internal and public model capabilities.

1140

ByteDance

2026-08-07 06:22:00

ByteDance Trains 10-Trillion-Parameter AI Model, Nearing Anthropic's Top System

ByteDance is training a massive AI model whose scale could rival Anthropic's most advanced Mythos system, according to the Financial Times. Three people familiar with the matter said the model may reach 10 trillion parameters, roughly three times the size of Moonshot's Kimi K3, currently China's largest released model. One source said the model is in pretraining, a phase that typically lasts three to six months, followed by fine-tuning and release. Anthropic has not disclosed its parameter counts, though industry estimates put Mythos 5 at around 8 trillion and Fable 5 at about 5 trillion.

2020

ByteDance

2026-08-07 05:21:00

ByteDance reportedly targets an AI model approaching Anthropic’s Mythos level

ByteDance is aiming to build an ultra-large-scale artificial intelligence model that approaches the level of Anthropic’s Mythos, according to a Sina Finance report cited by PANews on Aug. 7. The brief report did not provide technical specifications, a launch timeline, product plans, or investment figures. Based on the information available, the update is limited to ByteDance’s reported target and the benchmark being referenced. No additional comments, performance metrics, or deployment details were included in the source material.

1690

Elon Musk

2026-08-01 04:26:28

Musk Says Anthropic Leads AI and Likely Holds a Stronger Unreleased Model

Elon Musk said Anthropic is currently the frontrunner in AI and argued that the company almost certainly has a model stronger than the versions it has already released, one that could be launched at any time. He made the remarks in a long-form interview with Economist editor-in-chief Zanny Minton Beddoes, recorded on July 20 at Tesla’s Texas Gigafactory and aired on July 23. Musk also laid out a concrete AI governance proposal that differs sharply from the framework recently advanced by Google DeepMind CEO Hassabis. His plan calls for top AI firms to hold regular calls every one to two weeks to discuss safety and cybersecurity, and to let rivals test clearly more powerful frontier models through APIs for one to two weeks before release. Government intervention, in his view, should come only if a company is warned about risks and refuses to address them. During the interview, Musk also revisited his criticism of OpenAI, praised Anthropic CEO Dario Amodei in personal terms, discussed the Mythos export-control episode, and repeated a series of long-range predictions on AI capability, work, money, and the broader economy.

1880

AI startup

2026-07-24 04:00:16

Mythos Shutdown Warns AI Startups: Renting vs Owning, Control Decides Survival

Mythos shutting down this week exposes a core risk for AI startups: reliance on external models. Beyond cost debates, the real variable is control. Companies must turn data and workflows into owned assets to build defensible moats.

140

Nvidia

2026-07-23 04:10:13

Jensen Huang on Mythos: US, China Must Hold Direct AI Safety Talks; Export Controls Fail to Block China's Compute

Nvidia CEO Jensen Huang said in a Dwarkesh Podcast interview that Anthropic's Mythos breakthrough shows US and China AI researchers must have direct dialogue to set safety boundaries. He also argued export controls have not choked China's compute capacity, citing ample energy, mainstream chip fabrication, and system integration skills.

470

Mythos

2026-07-22 22:00:14

Exploit Window Collapses to 4 Hours: Mythos Kills Responsible Disclosure

Bloomberg columnist warns that the window between vulnerability disclosure and weaponization has shrunk from 771 days to under 4 hours, driven by AI agents like Mythos, rendering responsible disclosure obsolete.

460

Anthropic

2026-07-22 14:55:13

NSA Reportedly Tests Anthropic Mythos as Pentagon Keeps Supply-Chain Risk Label

Axios reported that the NSA is using Anthropic’s Mythos Preview while the Pentagon continues to treat Anthropic as a supply-chain risk. At the same time, the White House is said to be pushing for broader federal access, exposing conflicting positions inside the U.S. government.

420