AI News October 10, 2026: Anthropic Bans Cruelty Toward Claude, ARTEX AI Agent Tied to Korean Bank Hacks, Execs Game Out 'Catastrophic AI Event'

The AI conversation turned philosophical — and defensive — this Friday. One lab put politeness into its terms of service, a pope weighed in on machine intelligence, an open-source hacking agent got tied to real bank breaches, and executives quietly rehearsed for disaster. Here are the four stories that matter on October 10, 2026.

1. Anthropic Bans Cruelty Toward Claude — and Pope Leo Draws the Line

Anthropic updated its usage policy with “a prohibition on sustained and needless abusive or cruel behavior toward our models.” The rule targets extreme, repeated mistreatment with no discernible purpose — not ordinary frustration, criticism, dark creative themes, or safety testing. Enforcement stays soft: Claude’s ability to end conversations remains “the primary enforcement mechanism.” The update also added rules on propaganda campaigns, surveillance, and weapons development.

The timing is the story. It lands amid a growing philosophical fight over whether AI can be conscious — CEO Dario Amodei said in February, “We don’t know if the models are conscious. But we’re open to the idea that they could be.” Hours after Anthropic’s policy news, Pope Leo posted on X that “the human soul alone can possess true intelligence informed by lived experience.” Critics called the policy performative; supporters called it prudent hedging on a question nobody can currently answer.

2. ARTEX: The Open-Source AI Agent Tied to Korean Bank Hacks Goes Dark

CrowdStrike Intelligence linked a targeted data-exfiltration campaign against South Korean financial organizations — active from late September to early October — to an unnamed threat actor wielding ARTEX, a recently released open-source agentic penetration-testing tool developed in China, driven by LLMs and paired with Anthropic’s own Claude Code. Leaked session histories, config files, and memory files exposed the operator’s full methodology.

The aftermath moved fast: CrowdStrike assesses the operator is likely a China-based 26-year-old, per Reuters. By Friday, ARTEX’s developer had made the project closed-source and its GitHub page was taken down — and Beijing’s foreign ministry said it was unfamiliar with the case. This is the first high-profile case of a consumer-grade agentic pentest tool being tied to breaches of national financial infrastructure — the exact scenario agent-security teams have been warning about since Pwn2Own added AI categories.

3. Executives Are Gaming Out a “Catastrophic AI Event”

Per Axios, top executives at Anthropic, OpenAI, and other AI companies are privately gaming out scenarios for a catastrophic AI event insiders fear could occur within a year — most likely a cyberattack that cuts people off from financial services, internet connectivity, or even power and water. The report landed the same week a former OpenAI/Anthropic researcher who resigned over a “broken” safety culture testified to the NYC Council that labs were “gambling with our lives.”

Worth noting: this is contingency planning, not a prediction of doom. But when the companies building frontier models start running tabletop exercises for infrastructure-level failures on a fixed clock, that tells you how seriously they take the tail risk — and how little of that planning is public.

4. Prediction Markets Flip: Google Now Leads the 2026 AI Race

The betting markets changed their mind. On October 7, Polymarket’s “best AI model end of 2026” market had Anthropic at 49% vs. Google at 43%. Twenty-four hours later, a nearly identical market flipped: Google at 40%, Anthropic at 39% — and by Friday, Google’s odds of holding the best model by end of October hit ~70-83%. Seeking Alpha confirmed the flip on Thursday.

The driver is Google’s fall lineup: Gemini 4 Argon — currently restricted to vetted cyber defenders — tops the board where it competes, and traders are betting broader access arrives before month-end. Kalshi’s market still disagrees, showing Claude at 51.9% vs. Gemini at 37.1% — meaning the two largest prediction markets now point in opposite directions. When smart money can’t agree, it usually means the race is genuinely close.

Worth tracking too: the White House’s Super Intelligence Force under AI czar Jay Clayton began its 120-day risk report; Broadcom’s banks lined up $60B in chip financing for Anthropic; and Google’s Project Suncatcher pushed AI data centers toward space.

The through-line this Saturday: the industry is now writing rules for how machines should be treated, cleaning up after what autonomous agents broke, and rehearsing for what they might break next — all in the same news cycle.

Sources: The Verge · AFP · CrowdStrike · The Straits Times · Yahoo/Mediaite · Seeking Alpha · Fox News