AI News Digest, July 30: South Korea Just Showed What Real AI Compliance Looks Like

South Korea flipped a switch this week that most companies hiring globally haven’t noticed yet. Its revised Korea AI Basic Act enforcement decree took effect July 21. It tightens rules for any AI vendor selling into the country. But regulators are still keeping most fines paused for another year. That combination is unusual: real, binding law, paired with a deliberate grace period. Meanwhile, three other stories from the past 48 hours point the same direction. HR teams are already using AI for jobs that aren’t theirs. A Spanish startup just raised $570 million to make AI cheaper to run. And the protocol connecting AI agents to company systems got its biggest rewrite yet. Here’s what changed and what to do about it.

South Korea’s Revised AI Basic Act Enforcement Decree Takes Effect

South Korea approved a revised enforcement decree for its Korea AI Basic Act on July 14. The update took effect July 21, tightening the rules for AI vendors that sell into the country. (Source: MLex) The original law took effect January 22. That made South Korea the second jurisdiction in the world, after the EU AI Act, to enforce a comprehensive binding AI framework. The revision adds a wrinkle that matters if you sell software into Korea. Any AI company must now designate a Korean legal representative once it crosses $662 million in prior-year total revenue, or $6.6 million in Korean AI-service revenue, or one million average daily Korean users over a trailing three-month period.

What the Korea AI Basic Act Means for Global HR Teams

If your HR stack includes AI screening, scheduling, or payroll tools with a meaningful Korean user base, check whether your vendor has crossed that threshold. Most HR SaaS companies haven’t hit $6.6 million in Korean AI revenue yet, so the representative requirement won’t bite most of you directly. But the government is also giving procurement preference to domestic AI products. It’s also refining who qualifies as an “AI-vulnerable group” eligible for support. Both are signs of a market that’s about to get more protective of local vendors. At the same time, regulators are holding off on enforcement. MSIT is running a grace period through the rest of 2026, deferring fines except in cases of serious harm. So foreign vendors get a real runway before penalties start. Companies hiring engineers or expanding operations in Seoul should treat this as the moment to get ahead of it, not wait for the grace period to end. Asanify’s Employer of Record in South Korea already tracks regulatory shifts like this one as part of onboarding. That way, you don’t find out about a new representative requirement the hard way.

HR Teams Are Quietly Using ChatGPT to Do Marketing’s Job

OpenAI analyzed more than 800,000 ChatGPT messages. It found that 44% of all occupation-specific requests involved tasks outside the user’s actual job. Among HR professionals, that share jumped to 69%, the third-highest rate of any occupation studied. (Source: HR Dive) HR practitioners spent that borrowed time mostly on marketing tasks, 23% of their messages. Engineering came next at 18%, then finance at 16%.

So what? Maybe your HR team is already writing marketing copy. Maybe they’re building the kind of spreadsheet formulas that used to require a finance analyst. Either way, the org chart is quietly changing shape before anyone updates the job descriptions. That’s worth raising at your next leadership meeting. It affects headcount planning more than most new HR software ever does. The AI skills gap in HR isn’t only about HR learning new tools. It’s about HR absorbing work that used to belong to other departments.

A Spanish Startup Just Raised $570M to Make AI Cheaper to Run

Multiverse Computing is raising up to $570 million in a Series C round. That would value the company at $1.7 billion, roughly a five-times step-up from its Series B. (Source: Bloomberg) The company’s CompactifAI technology uses tensor-network methods borrowed from quantum physics. It shrinks large language models by 80 to 95%, with little accuracy loss, so they run cheaper on ordinary hardware.

So what? Inference cost is the line item most founders underestimate when they budget for AI features. Model-compression companies like Multiverse keep raising at this pace. If that continues, expect the AI tools you already use to get noticeably cheaper to run within the next year. Not because vendors got generous, but because the underlying compute got lighter. Worth watching if your product roadmap has AI features you’ve held back on for cost reasons.

The Protocol Behind AI Agents Just Got Its Biggest Rewrite Ever

The Model Context Protocol’s July 28 specification update is its largest revision since launch. It removes the old session-based handshake and adds a stateless core built for standard HTTP scaling. It also aligns authorization with OAuth 2.1 and OpenID Connect, so enterprises can manage AI agent access centrally. (Source: The Register) MCP is the standard a growing number of AI agents use to connect to company tools and data.

So what? Maybe your engineering team has held off connecting AI agents to internal systems, because the security story felt half-built. This update is the fix. Centralized authorization means one login can now govern which AI agents can touch which systems. Before, every integration invented its own access rules. That’s the difference between a pilot project and something IT will actually approve for production.

Quick Hits

• Act Security emerged from stealth with $60 million to fix the AI-era patch problem. Its argument: cloud access sprawl, not unpatched software, is now the bigger risk. It’s the kind of gap a solid cybersecurity policy is supposed to close. (SecurityWeek)
• Malaysia’s public consultation on its first horizontal AI Governance Bill closes July 31. The risk-based framework would add incident-reporting duties and regulatory sandboxes across the AI lifecycle. (Digital Watch Observatory)
• 43% of managers say they’re poorly equipped, or not equipped at all, to lead an AI-fluent workforce. That’s a new Indeed and YouGov survey finding. Employer expectations for advanced AI use still run two to three times ahead of what workers feel ready for. (HR Dive)

Today’s Korea AI Basic Act news might have you wondering what other countries are quietly tightening their AI and employment rules. That’s exactly the kind of shift Asanify tracks in every market where we help you hire. Worth a look before your next hire in a new country catches you off guard.

FAQ

What does the Korea AI Basic Act enforcement decree change?

The revised decree took effect July 21, 2026. It requires foreign AI companies above certain revenue and user thresholds to appoint a Korean legal representative. It also expands procurement preference for domestic AI products and refines support eligibility for AI-vulnerable groups. Regulators are keeping most fines paused through a 2026 grace period, except in cases of serious harm.

Is South Korea’s AI law the first in Asia?

It’s the first comprehensive, legally binding AI framework in Asia. It’s only the second in the world, after the EU AI Act. The Korea AI Basic Act and its enforcement decree took effect January 22, 2026. That’s ahead of similar efforts like Malaysia’s AI Governance Bill, which is still at the public consultation stage.

Why are HR professionals using ChatGPT for non-HR work?

OpenAI’s analysis of more than 800,000 messages found HR professionals cross into other departments’ work in 69% of their occupation-specific ChatGPT use. That’s mostly marketing, engineering, and finance work. It’s the third-highest rate of any profession studied, and it suggests AI is blurring job boundaries faster than org charts are getting updated.

Not to be considered as tax, legal, financial or HR advice. Regulations change over time so please consult a lawyer, accountant  or Labour Law  expert for specific guidance.