Skip to main content

8 min read

News date: September 30, 2026.

On September 30, 2026, Google announced Gemini 4 Argon, which it calls a new frontier AI model built for “deep reasoning across complex, long-horizon workflows.” What makes this launch unusual is who gets it first. Instead of releasing Argon to everyone, Google is rolling it out to “a set of trusted cyber defenders” through its Fairwind Program. Google says Argon can autonomously find, validate and patch critical software vulnerabilities, and it wants defenders to have that capability before wider release.

For Philippine businesses, the launch is a sign of where cybersecurity is heading. AI is becoming very good at finding weaknesses in software, which helps defenders but also raises the stakes for anyone running outdated or poorly maintained websites and apps. This article explains what Google announced and what it means locally.

The key facts

  • What it is: Google’s new frontier model, which Google says delivers leading performance in real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence.
  • Who can use it now: trusted cyber defenders through Google DeepMind’s Fairwind Program. Google says it will continue to gather feedback and improve guardrails “before making Argon available to developers, enterprises, and consumers as soon as possible.”
  • Output limit: Google says it is expanding Argon’s output token limit to 1 million tokens, up from 64,000, to support long, complex tasks.
  • Pricing: Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input priced at 95% off. After the introductory period, $4 per million input tokens and $20 per million output tokens will apply.
  • Safety approach: Google says it is taking part in the U.S. government’s voluntary pre-release model access process and expanding access gradually.

Focus on cybersecurity defence

Google says it trained Argon to be highly capable at cybersecurity defence. For trusted defenders and Google’s own teams, Google is releasing Argon without cyber guardrails so they can use its full defensive capabilities. Examples from the announcement include:

  • In an early demonstration, the model uncovered a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide, a risk Google says previous frontier models had missed.
  • On Google’s internal vulnerability benchmark, Argon found exposures across complex codebases in 20 programming languages.
  • On a black-box penetration testing benchmark from security company Wiz, which tests analysis of live web systems without source code, Google says Argon outperformed its previous cyber-focused model in discovering the attack surface, identifying vulnerabilities and producing proof-of-concept evidence.

According to Google DeepMind’s Fairwind pages and an analysis by WorkOS, Fairwind access is for vetted organisations such as governments, critical infrastructure operators, healthcare providers and telecoms. Requirements include organisational vetting, phishing-resistant multi-factor authentication, access limited to security, incident response or penetration testing teams, and tracked employee use. Access cannot be shared, resold or redistributed.

Other capabilities Google highlighted

Google says thousands of its employees already use Argon internally. Examples in the announcement include agents that analysed profiling data across Google’s data centres and applied memory optimisations, and large-scale migration of C and C++ code to the memory-safe Rust language, with automated and manual auditing before production. Google reports a score of 77.9% on DeepSWE v1.1, a benchmark of long-horizon software engineering tasks, and strong results on finance, legal and long-video understanding benchmarks. These are Google’s own reported results.

Google also says Argon is its most resilient model yet against indirect prompt injection, where malicious instructions hidden in content try to hijack an AI’s behaviour, and that it is deploying monitoring to stop the model from acting beyond the user’s intentions.

Why this matters: the speed of finding vulnerabilities

The important point for business owners is not the model’s name. It is the trend. AI systems are becoming able to scan code and live websites, find weaknesses and prove they can be exploited, faster than human teams. Google’s approach is to give this ability to defenders first, but it is reasonable to expect attackers to use increasingly capable tools as well.

That shortens the time between a vulnerability becoming known and it being exploited. A WordPress plugin with a known security flaw, an unpatched server or an exposed admin panel will be found more quickly than before.

It is also worth remembering that most successful attacks on small businesses are not sophisticated. They exploit an old plugin, a reused password or a phishing email. Advanced AI raises the ceiling of what attackers can do, but the floor, basic hygiene, still prevents most incidents. Businesses that get the basics right are in a far stronger position whatever tools attackers use.

Illustration of an AI scanner rapidly finding weak points in website code
AI tools can scan code for weak points far faster than manual review.

What this means for Philippine businesses

Recent incidents involving government systems in the Philippines, including the LTFRB’s records platform in September 2026, show that data breaches are a real local risk. Most small and medium businesses will not have access to Argon, but they can act on the lesson.

1. Patch faster

Treat security updates for your website, app, server and plugins as urgent. Set a routine, such as weekly checks, with immediate action for critical updates.

2. Reduce your attack surface

Remove unused plugins, old test sites, forgotten subdomains and admin tools that are open to the internet. Every unnecessary component is another possible entry point.

3. Use phishing-resistant login for admins

Fairwind itself requires phishing-resistant multi-factor authentication. Businesses can adopt the same principle: use passkeys or security keys where available, or at least app-based two-factor authentication, for every admin and hosting account.

4. Test your own systems

Arrange periodic security checks or penetration tests for important websites and apps, especially those handling payments or personal data. Fix the findings and retest.

5. Prepare for AI features securely

If you add AI assistants to your website or app, consider prompt injection. Do not let an AI assistant take actions, such as refunds or account changes, without proper checks, and limit the data it can access.

6. Keep good backups and a response plan

Even well-protected systems can be breached. Tested backups and a simple incident response plan, including notification duties under the Data Privacy Act, reduce the damage.

Common weaknesses that automated tools find quickly

Automated scanners, with or without AI, tend to find the same kinds of problems on small business websites and apps. These are worth checking now:

  • Outdated components: old versions of WordPress, plugins, themes, PHP or JavaScript libraries with publicly known vulnerabilities.
  • Weak access control in APIs: mobile app backends where changing an order or user ID in a request shows someone else’s data. This is one of the most common and serious problems in custom apps.
  • Exposed files: backups, database exports, configuration files or uploaded IDs stored in public folders.
  • Default or reused passwords on admin panels, databases and hosting control panels.
  • Missing rate limits: login and OTP endpoints that allow unlimited guessing.
  • Verbose error messages that reveal file paths, database details or software versions.

Fixing these basics removes much of the risk, because they are exactly what automated tools look for first.

When Argon reaches developers

Google has said it plans to make Argon available to developers, enterprises and consumers after more testing, without giving a date. When that happens, development teams may be able to use such models to review code for security issues before release, in addition to human review and established security tools. Businesses commissioning software can then ask their developers how AI-assisted code review and security testing fit into the development process, while remembering that automated findings still need expert verification.

A quick security checklist

  • All software updated within the last month, with critical patches applied immediately.
  • Unused plugins, themes, test sites and accounts removed.
  • Two-factor or passkey login on all admin, hosting, domain and email accounts.
  • Web application firewall and malware scanning enabled.
  • Daily off-site backups, with a restore tested in the last three months.
  • Logs of admin logins and changes kept and reviewed.
  • A named person responsible for security and breach response.
Illustration of a website security checklist with updates, backups and two-factor login
Updates, backups and two-factor logins close the gaps automated scanners find first.

Frequently asked questions

Do small business websites really get attacked?

Yes. Most attacks are automated. Bots scan the internet for known weaknesses, such as outdated plugins, and do not care how big your business is.

What is the single most useful step?

Keep everything updated: your CMS, plugins, themes and server software. Many successful attacks use vulnerabilities for which a fix already exists.

Can AI tools scan my site for me?

AI-assisted tools can help developers review code and configuration faster, but results need a person to check them. Only scan systems you own or have written permission to test.

What laws apply if my site is hacked?

If personal data is exposed, the Data Privacy Act of 2012 and National Privacy Commission rules on breach notification apply. Online fraud and illegal access are also covered by the Cybercrime Prevention Act of 2012 (Republic Act No. 10175), and incidents can be reported to the PNP Anti-Cybercrime Group or the CICC.

Further reading on ServicioPro

Read our guides on how to choose a web developer in the Philippines and app privacy compliance basics for more on building secure, privacy-conscious systems.

References

  • Google, “Introducing Gemini 4 Argon,” September 30, 2026. blog.google
  • Google DeepMind, “Fairwind Program.” deepmind.google
  • WorkOS, “Gemini 4 Argon access: What Google’s Fairwind rules require.” workos.com