Can You Use ChatGPT in Hong Kong? AI Tools That Work in 2026

Hong Kong’s access to mainstream AI has been a moving target — and in 2026 it shifted again. The headline change: Google Gemini is now officially available to everyone in Hong Kong, the first major US consumer assistant to drop its HK exclusion. ChatGPT and Claude still aren’t directly available, but the models behind them are reachable through other routes. This guide gives you the current picture (as at May 2026 — this space moves fast): what works, how to use it safely, and what a Hong Kong business should actually deploy.

One thing to clear up first, because it drives everything else: Hong Kong’s internet is open and uncensored. There is no Great Firewall here and no government block on these tools. Where a Western assistant is unavailable, it’s the vendor geofencing Hong Kong as a commercial decision — not a state block. That distinction matters, and the gap is closing.

Where Hong Kong’s AI access stands now (mid-2026)

• Google Gemini is now available — officially, to all HK users, since March 2026. Free tier plus paid via Google One. This is the big change.
• ChatGPT is still blocked directly by OpenAI — but you can use the same GPT models through Microsoft Copilot (the compliant enterprise route) or Perplexity.
• Microsoft Copilot is fully available and runs OpenAI’s models inside Microsoft 365. For most businesses, this is the answer.
• Anthropic Claude is not available directly (claude.ai blocks HK), but is reachable indirectly via Perplexity.
• Perplexity is available and mainstream — HKT even bundled free Perplexity Pro to consumers in 2025 — and it routes to GPT, Claude, Gemini and others.
• DeepSeek and other China-developed models are freely available, Cantonese-capable, and increasingly relevant for cross-border work.
• Grok (xAI) is available without restriction, though thin as an enterprise tool.

At a glance: which AI tools work in Hong Kong

Tool	Available to HK users?	How to use it	Notes
Google Gemini	✅ Yes — since March 2026	Direct: free tier + Google One paid	First major US assistant to open to HK
Microsoft Copilot	✅ Yes	Direct, inside Microsoft 365	Runs OpenAI’s models; the compliant enterprise route
ChatGPT (OpenAI)	❌ Not directly	Via Copilot or Perplexity	OpenAI geoblocks HK IPs / +852 numbers — its choice, not a HK block
Anthropic Claude	❌ Not directly	Via Perplexity / aggregators	claude.ai blocks HK; the model is still reachable indirectly
Perplexity	✅ Yes	Direct	Routes to GPT, Claude, Gemini, DeepSeek and more
Grok (xAI)	✅ Yes	Direct	Thin for enterprise vs Copilot
DeepSeek	✅ Yes	Direct: free, Cantonese-capable	China-developed — apply data-governance caution for regulated data

ChatGPT is still blocked — but the model isn’t out of reach

In mid-2024, OpenAI restricted ChatGPT and its API across a list of “unsupported regions” that includes Hong Kong, and that restriction still stands — Hong Kong does not appear on OpenAI’s supported-countries list. ChatGPT.com returns errors from HK IPs, and API calls from +852-registered accounts are cut off.

But “the app is blocked” is not “you can’t use the model.” The same GPT models are available, compliantly, through Microsoft Copilot (below), and reachable for casual use through Perplexity. So the practical answer for a Hong Kong business has never really been “do without GPT” — it’s “use it through the right door.”

Apple Intelligence on Hong Kong iPhones

Apple Intelligence combines on-device models with a private cloud layer and, where the user opts in, hands off to ChatGPT via Siri. The on-device layer works fine in Hong Kong — summaries, Genmoji, notification digests, and local Writing Tools all function. The ChatGPT integration, however, still fails, because it relies on OpenAI’s infrastructure, which blocks Hong Kong. The toggle appears under Settings → Apple Intelligence & Siri, but enabling it returns “ChatGPT with Siri is not supported on this Apple Account.”

For organisations issuing iPhones, the implication is unchanged: on-device features work, but the broken ChatGPT affordance pushes staff toward VPNs and unsanctioned apps. Apple Business Manager plus MDM can disable the option cleanly and steer people to a sanctioned route — exactly the kind of device policy we set up as part of managed IT.

Microsoft Copilot: the compliant enterprise route

Microsoft Copilot is fully supported in Hong Kong — Microsoft’s own supported-regions list explicitly carves Hong Kong out of the China exclusion (“China (excluding Hong Kong)”). It runs the same OpenAI GPT models that ChatGPT does, so the contradiction works in business users’ favour. What it gives you:

• Enterprise integration inside Word, Excel, Outlook, Teams and the rest of Microsoft 365.
• Security and compliance under Microsoft’s enterprise data-protection commitments — prompts and responses aren’t used to train the models, and data stays within the tenant’s boundaries.
• Audit and governance through standard M365 admin tooling — DLP, sensitivity labels and retention policies apply to Copilot output.

For most Hong Kong businesses already on Microsoft 365, this is the right answer — the only OpenAI-backed service that can be deployed legally, compliantly and at scale here. The trade-offs are deeper Microsoft lock-in and a higher per-user cost than a personal ChatGPT subscription. Treat a rollout as a real change programme, not a switch you flip — see Microsoft 365 setup and Microsoft 365 security, and our Microsoft 365 in China guide for the tenant-choice considerations.

The other models: Gemini, Claude, Perplexity, Grok and DeepSeek

Google Gemini is the big 2026 change — now officially available in Hong Kong on a personal Google account, free tier plus paid Google AI Pro. No VPN, no workaround. For the first time, a major US consumer assistant is open to HK as standard.

Anthropic Claude remains unavailable directly — Hong Kong isn’t on Anthropic’s supported-countries list and claude.ai blocks HK IPs. The model is still reachable through Perplexity, which exposes Claude as one of its selectable models.

Perplexity is fully available and has gone mainstream — and because it routes queries to a mix of models (GPT, Claude, Gemini, DeepSeek and its own Sonar), it’s how many HK users reach capability that the underlying vendors’ own apps still block. Useful — but for regulated or client data, an aggregator with unclear data residency is the wrong place to put it.

Grok (xAI) is available without restriction; capable as a conversational assistant, thin as an enterprise platform.

DeepSeek and other China-developed models (Alibaba’s Qwen, others) are freely available in Hong Kong, handle Cantonese well, and broke out here in 2025. For businesses with Mainland operations they’re worth understanding — but mainland-hosted models carry their own data-governance questions, which run straight into China’s data laws. See also IT services in China.

The rules are catching up: Hong Kong’s AI governance

This is the part that’s changed most since this post first ran — Hong Kong now has concrete published guidance, not just consultations:

• PCPD — Model Personal Data Protection Framework for AI (June 2024) and a GenAI-for-employees checklist (March 2025): how to procure and use AI involving personal data, and how to set an internal staff-use policy. (PCPD AI guidance)
• Digital Policy Office — Ethical AI Framework (2024) and the Hong Kong Generative AI Technical and Application Guideline (April 2025): operational guidance on data leakage, bias and accuracy.
• For financial firms, the SFC’s circular on generative AI (November 2024) sets expectations for licensed corporations using GenAI in regulated activities — covered in our guide to SFC cybersecurity requirements for Hong Kong fund managers.

The common thread: regulators don’t ban these tools, but they expect you to have an AI acceptable-use policy and to not feed personal, confidential or client data into public AI tools. That sits alongside your existing obligations under the PDPO and, for cross-border data, PIPL.

What Hong Kong businesses should do

1. Stop shadow AI. Don’t let staff route business data through VPNs or unverified third-party apps — for financial, legal or healthcare firms it can put you in breach. Give them a sanctioned tool instead.
2. Pick your sanctioned route. For Microsoft 365 shops, Copilot is the cleanest. Gemini and Perplexity are now legitimately available too — but apply governance before regulated data goes near them.
3. Write an AI acceptable-use policy and back it with technical controls — DLP, sensitivity labels, retention — aligned to the PCPD and DPO guidance above. ISO 27001 clauses A.12.6 and A.13.2 are useful anchors. This is core cybersecurity work.
4. Manage Apple devices with MDM to remove broken integrations and prevent the VPN scramble.
5. Financial firms: meet the SFC’s GenAI expectations before staff start using AI in regulated work.
6. Cross-border? Understand the China picture — DeepSeek and the data-residency rules — before you deploy anything that touches the Mainland.

How PTS helps

PTS helps Hong Kong businesses adopt AI safely: Microsoft Copilot rollout and licensing, AI acceptable-use policy and governance, DLP and data classification, Apple/MDM device policy, and the cross-border considerations for firms with Mainland operations. We work to ISO/IEC 27001 and alongside your compliance and legal advisers — so you get the productivity without the shadow-IT and data-protection risk.

If you’re planning AI adoption — or just want a starter AI acceptable-use policy for your team — talk to PTS and we’ll come back with a practical, costed plan.

Related reading: AI tools compared — what works in Hong Kong & China · Practical AI advisory for Hong Kong & China · Microsoft AI in Hong Kong schools · Microsoft 365 in China · data laws in China & Hong Kong · SFC cybersecurity for fund managers · AI scams in Hong Kong · 19 cybersecurity policies every business needs