The AI Brief

On May 14, 2026, Governor Jared Polis signed SB 26-189, which repeals and replaces Colorado's original AI Act (SB 24-205). The new law delays the effective date from June 30, 2026, to January 1, 2027, while significantly scaling back the original requirements. It eliminates the duty of care aimed at preventing algorithmic discrimination, deployer obligations to maintain risk management programs and conduct impact assessments, and certain reporting obligations to the Colorado Attorney General, adopting instead a narrower approach focused on disclosures and transparency around automated decision-making technologies.

The path to this outcome was unusually compressed. The Senate passed SB 26-189 on May 7 by 8-1, the House passed it on May 9, and Polis signed it May 14. Before the legislature acted, enforcement had already been frozen: on April 27, 2026, a federal magistrate judge stayed enforcement of the Colorado AI law. xAI had filed suit challenging the law on constitutional grounds on April 9, 2026, and the DOJ moved to intervene on April 24 to join xAI's effort to invalidate the law, raising additional constitutional challenges. The court stay technically extends to SB 26-189 as well, meaning enforcement of the successor law also awaits completion of the attorney general's rulemaking process, which has not formally started.

Colorado was the bellwether for state AI regulation aligned with the EU model. Its quick about-face, executed with weeks remaining before the original law's June 30, 2026 effective date and amid active federal pressure on the same statute, is the strongest signal yet that the EU template will not be the dominant US state framework. Whether the replacement model converges nationally or fragments into a patchwork like the existing state-by-state privacy regime remains to be seen. Multinationals with US and EU exposure will increasingly need to maintain two distinct compliance postures rather than one harmonized program.

80% of enterprise applications shipped or updated in Q1 2026 embed at least one AI agent, per Gartner, up from 33% in 2024. The Gartner projection for year-end 2026 places the share of applications with task-specific agents at 40%, reflecting the difference between any embedded agent (the 80% figure) and agents scoped and tested for a defined production task. The 2024-to-2026 jump is steeper than any comparable enterprise software adoption curve since cloud computing in 2010-2012.

31% of enterprises have at least one AI agent in production, per S&P Global Market Intelligence and McKinsey, with banking and insurance leading at 47% and healthcare and government trailing at 18% and 14% respectively. 22% of production deployments now coordinate three or more agents, and 56% of enterprises now name a dedicated AI agent owner or agentic ops lead in 2026, up from 11% in 2024.

CaveatThe 40% figure is a Gartner forward projection, not a measured outcome. The 80% figure for Q1 2026 embeds agents across a wide capability range, from simple deflection bots to autonomous coding agents. All figures are from analyst surveys with vendor-interest dynamics; treat them as directional rather than precise.

In January 2026, METR released Time Horizon 1.1. According to the updated model, the rate of progress of AI capabilities has increased since 2023, with a post-2023 doubling time estimated at 130.8 days (4.3 months). Progress is thus estimated to be approximately 20% more rapid than the longer-run trend.

The original METR paper, published in March 2025, measured the doubling time at approximately seven months across 2019-2024. In 2024-2025, time horizons doubled every four months, down from every seven months over 2019-2025. If the faster trend continues, agents might reach month-long tasks in 2027. However, looking at just one year's data gives a less robust estimate. The rate of progress might slow down. It might also speed up.

METR added Claude Mythos Preview to its published time-horizon table on May 8, 2026, along with a notice that measurements above 16 hours are unreliable with its current task suite. The 16-hour ceiling is a methodological constraint rather than a capability ceiling: it reflects the limits of METR's current benchmark suite, not the outer boundary of what frontier agents can attempt. METR is a nonprofit that has worked with OpenAI and Anthropic on pre-deployment evaluations.

NoteMETR's task suite focuses on software-engineering and adjacent tasks; generalization to other domains is unverified. The January 2026 estimate uses bootstrapped confidence intervals; the post-2023 subset has a shorter observation window than the full 2019-2025 trend.

The lawsuit, filed Thursday in the U.S. District Court for the Southern District of New York, accused Perplexity of scraping more than 17,000 CNN stories, photos, videos and other content and using that to train its products. It is CNN's first AI copyright action and is thought to be the first by any television network. The filing indicates that CNN sought to strike a content deal with Perplexity last year but did not agree on terms.

Perplexity's response followed its established pattern. "You can't copyright facts," Perplexity's chief communications officer Jesse Dwyer said in a statement to CNN. Nine organizations have active suits against Perplexity for alleged copyright and trademark infringement as of May 31, 2026: CNN, the New York Times, News Corp and Dow Jones, the New York Post, the Chicago Tribune, Encyclopedia Britannica, Merriam-Webster, Reddit, and Japan's Yomiuri Shimbun. Other publishers, including Time, Gannett, Le Monde, and Der Spiegel, have signed licensing agreements with Perplexity rather than litigate.

The broader industry benchmark was set in August 2025, when Anthropic agreed to pay $1.5 billion to resolve the Bartz v. Anthropic class action, in which authors alleged the company had downloaded their books from pirate libraries to train its Claude models, the largest copyright settlement in US history. A fairness hearing was held on May 14, 2026, before Judge Araceli Martínez-Olguín in the Northern District of California; final approval had not been issued as of May 31.

Per Anthropic's official help center: "Starting June 15, 2026, Claude Agent SDK and claude -p usage no longer counts toward your Claude plan's usage limits." That usage draws from a new, separate Agent SDK monthly credit denominated in dollars and billed at standard Anthropic API rates. The split creates two independent buckets: the interactive pool (unchanged) covers Claude.ai web, desktop, and mobile chat, Claude Code used interactively in the terminal, and Claude Cowork. The Agent SDK credit pool funds programmatic and autonomous usage, billed at full API list prices, with no rollover.

For teams running production automation on Claude subscriptions, this is the most significant billing change since Claude Code launched. Claude subscriptions previously subsidized agent usage at roughly 15-30x compared to API pricing, and the new credits are billed at full API rates. The per-user, non-pooled nature of the credit is the underreported wrinkle. Teams running shared CI/CD pipelines, where a single GitHub Actions workflow might trigger under multiple committers' credentials, cannot aggregate credits across the team. For production shared automation, Anthropic's own documentation is explicit: "Teams running shared production automation should use Claude Platform with an API key for predictable pay-as-you-go billing."

GitHub, meanwhile, is transitioning Copilot toward a token- and credit-based system closely resembling Anthropic's latest changes. Analysts expect more vendors to create separate consumption pools for agents, premium models, tool use, and background tasks over the next 12 to 24 months.

Disclosure: Anthropic, mentioned in this item, is the company that develops Claude, which generates this brief.