AI Law Wiki News for September 15, 2026

Australia debates AI-training copyright options and protection for creators

On September 15, 2026, ABC reported previously undisclosed Australian copyright consultation options that would let rights holders opt out of AI training. One option would permit use of remaining online material after developers negotiated a minimum quota of licensing deals, without further payments beyond those deals. The attorney-general’s spokesperson said consultation continued on control and compensation; OpenAI confirmed participation. These were reported consultation options, not enacted changes.

In a same-day interview, Deputy Prime Minister Richard Marles disputed descriptions of OpenAI’s position as an investment ultimatum. He said Australia could attract AI training while protecting news organisations and creators, and acknowledged that copyright remained an issue under discussion.

South Korea prepares security guidance for autonomous AI agents

On September 15, 2026, South Korea’s Korea Internet & Security Agency (KISA) told Reuters it was revising its AI Security Guide to address autonomous agents. The planned update would include risk-management checklists and could cover AI operating physical equipment; it was described as work in progress, not released guidance.

KISA’s existing publication notice identifies the original guide as a December 10, 2025 resource for protecting AI models and services against cyber threats, with corrections issued on March 13, 2026.

Cloudflare introduces AI-training opt-outs that preserve search access

On September 15, 2026, Cloudflare announced “Disallow AI Training,” a setting that publishes no-training preferences while permitting search access by qualifying mixed-use crawlers. Its “Accountable” designation covers operators meeting, or committing to meet, requirements for training and AI-summary opt-outs, transparency and preserving traditional search ranking. This is a company content-use policy and technical control.

Cloudflare says the setting permits Applebot, Googlebot and Bingbot to continue search crawling, while blocking other training crawlers. It cautions that Bing’s automatic robots.txt training opt-out support is targeted for early 2027: the new setting does not yet communicate that preference to Bing, so separate controls remain necessary. Selecting “Block” instead can also block search crawling.

Google’s existing documentation separately confirms that Google-Extended controls specified Gemini training and grounding uses without affecting inclusion or ranking in Google Search.