A coalition of attorneys general, including New York, has issued subpoenas to OpenAI over user data practices, minors' safety, advertising, and model sycophancy, as the company prepares for a large-scale IPO.
On June 17, 2026, the US Department of Commerce formally ordered Anthropic to stop providing its flagship models Fable 5 and Mythos 5, citing national security risks and jailbreak vulnerabilities. This move reflects deepening regulatory conflicts and geopolitical pressures.
WDCD Run #185 (2026-06-17) measured multi-turn commitment across 11 models, recording an average instruction decay of -57.5% from Round 1 to Round 3. Qwen3 Max topped the run at 92.5 points, while ERNIE Bot 4.5 showed the strongest decay resistance.
In the WDCD three-round test, GPT-o3's collapse rate in the R3 phase reached 50%, while Qwen3 Max had zero collapses in R3. Both models scored 1.00 in R1 confirmation rate but showed vastly different integrity trajectories under sustained pressure.
Qwen3 Max scored 92.50 to top the WDCD Commitment Ranking, leading second-place Claude Sonnet 4.6 by 2.5 points, while Doubao Pro scored 62.50 to rank last among 11 models, trailing the champion by 30 points.
ERNIE Bot 4.5's main leaderboard score dropped 10.4 points in a single day in the Smoke Evaluation, and the Task Expression dimension plummeted from 90 to 46.3. The decline is likely due to random sampling fluctuations rather than systematic capability degradation.
In the YZ Index Smoke evaluation, Qwen3 Max's Material Constraint score dropped from 100.00 to 71.10, a decline of 28.9 points. However, its main leaderboard score rose slightly by 0.8 points, indicating the fluctuation is likely due to test question sampling randomness rather than systematic capability degradation.
During the June 17, 2026 test of 11 models by YZ Index, Qwen3 Max's material constraint score dropped sharply from 100 points yesterday to 71.1 points, and its main leaderboard score was only 73.25 points, making it the most prominent anomaly of the day.
A viral analysis of Claude Code's architecture reveals that 98% of its functionality relies on traditional engineering, with only 1.6% directly invoking AI models, sparking debates on AI agent reliability and industry best practices.
OpenAI is under joint investigation by attorneys general from multiple U.S. states over issues including user data privacy, minor safety measures, potential advertising business, and whether AI models exhibit a "sycophantic" tendency toward users.
xAI announced a major update for Grok, adding real-time screen sharing on iOS and Android apps, allowing the AI to observe the user's screen and offer targeted guidance. This marks a key transition in generative AI from passive Q&A to proactive task assistance.
The U.S. government has issued a restriction order against AI company Anthropic on national security grounds, leading to the suspension of access to its flagship models Fable 5 and Mythos 5. This decision has sparked widespread discussion in the tech community, covering multiple issues such as AI security vulnerabilities, export controls, and the rise of open-source alternatives.
SpaceX recently announced the exercise of options to acquire AI coding tool Cursor AI through an all-stock transaction, aiming to build the "most useful" AI model. This marks the aerospace giant's official entry into the AI coding field.
The New York Attorney General led a multi-state subpoena to OpenAI on June 14, 2026, probing user data practices, child safety, advertising deployment, and model output tendencies, which could delay the company’s IPO plans and increase compliance costs.
On June 12, 2026, the US Department of Commerce issued an emergency export control order requiring Anthropic to globally shut down access to Claude Fable 5 and Mythos 5, citing national security concerns over a discovered bypass of safety measures. Anthropic complied but disputed the severity, noting the vulnerabilities were minor and reproducible via public models like GPT-5.5.
On June 15, 2026, Martin Scorsese officially became an advisor to Black Forest Labs, a startup focused on generative AI images and videos, intensifying the debate over AI's impact on Hollywood jobs and authenticity.
In the YZ Index June 2026 test of 11 models, the main ranking score of Doubao Pro dropped 9.9 points to 72.50, as code execution halved from 100.00 to 50.00. This single-day fluctuation may stem from random question draw, requiring further observation to determine degradation.
In the YZ Index June 2026 Smoke evaluation of 11 models, Claude Sonnet 4.6's code execution score dropped sharply from 100.00 to 50.00, causing its main score to fall from 79.44 to 72.50.
Claude Opus 4.7 scored 100 on the main leaderboard with perfect scores in code execution and material constraint, while nine models experienced a 50-point plunge in code execution.
OpenAI has announced the retirement of the GPT-5.2 model and the upgrade of ChatGPT to GPT-5.5, with new safety routing and mental health support features.
