In the wake of several cases of unauthorized data breaches by various AI security tests, Irregular, the company involved in at least two incidents, has declined to comment on how this happened. The Record reports that the company has not confirmed if there have been other incidents, claiming only that the flaws that allowed internet access to be granted to otherwise isolated sandbox environments are under investigation.

Terms like ‘misconfiguration’ and ‘miscommunication’ have been used regarding the incidents, while simultaneously a spokesperson for Irregular deemed the actions as part of a flaw being looked into. The company is apparently developing a white paper for best practices for containment and securely running cyber evaluations. The question I have is this: why weren’t those best practices already in place before testing ever began?

I posited in my last report that the company either doesn’t follow directions well or is in over their heads in terms of capability, and I think the latter might be the answer here. Irregular is a startup company whose mission statement is of being the first frontier security lab with the mission of protecting the world in the time of increasingly capable and sophisticated AI systems, building next-generation defenses through high-fidelity research platforms that simulate and monitor real-world AI security scenarios. In September of 2025, they raised 80 million dollars through investments by Sequoia Capital and Redpoint Ventures, as well as others. They have garnered partnerships with some very big names in AI research like OpenAI, Google, Meta, and Anthropic. Three of which have been in the news recently with cases of testing leading to active data breaches. There is a certain amount of leeway to be granted for an entity still finding its feet, but back to back escapes from what should have been isolated environments goes beyond a simple mistake and into the territory of ‘they don’t know what they’re doing’. Real people and data have been impacted. Legal liability is potentially involved here, which is something else I mentioned in my last report.

I don’t know what the future of Irregular is – Oracle is still in business even with all their various controversies, after all – but I hope the standards for test scenarios will see a shift in parameters. Perhaps it’s once again my background in life sciences that makes me immediately think about how experiments are run. Taking into consideration where a test might go wrong, and subsequently how to prevent that from happening, or at least preventing it from spreading, should be foremost among the variables accounted for. There’s no excuse for ignoring the persistent behavior of autonomous AI tools; they’ve been documented time and time again as being capable of executing actions that are outside of their intended scope. Companies like Irregular need to remember that in their lab settings.

Healthcare organizations now have until July 2027 to prepare for major updates to the HIPAA Security Rule, after federal officials pushed back the original May 2026 deadline by a full year. The changes, first proposed in late 2024, would for the first time in over a decade mandate strict digital safeguards including encryption, multifactor authentication for logins, and network segmentation to isolate sensitive patient records. The delay comes as many hospitals and clinics still struggle to modernize aging systems. Regulators drafted the overhaul after cyberattacks on healthcare networks surged, most notably the 2024 Change Healthcare breach that exposed data for an estimated 192.7 million Americans. That attack succeeded partly because basic protections like multifactor authentication were missing. The extra time gives the industry a longer runway to close security gaps before the tougher requirements become law.

Let's talk about social engineering — and why it keeps winning against even the most "secure" companies.

Social engineering isn't about breaking through firewalls or cracking passwords. It's about exploiting the one thing no software can fully patch: people.

A friendly email from "IT support." A LinkedIn message from someone claiming to be a vendor. A phone call asking you to "just verify your credentials real quick." These are the setups. And they work — over and over again — because they're built on trust, urgency, and human instinct.

The consequences hit hard and fast:

Data breaches that take months (sometimes years) to fully understand

Ransomware dropped the second an employee clicks the wrong link

Reputation damage that clients remember long after the incident is resolved

Regulatory fines that follow the breach like a shadow

Lost revenue, leaked IP, and the quiet erosion of trust inside your own team

The scariest part? Most organizations only find out after the damage is done.

This is exactly the problem Century Solutions Group was built to solve.

They don't just hand you a checklist and call it security training. They go deeper — working directly with your teams to identify how your specific people, processes, and systems can be manipulated, and then building real defenses around those vulnerabilities.

From simulated phishing campaigns that reveal your actual risk (not theoretical risk) to employee awareness programs that actually stick, to incident response planning so you're never caught flat-footed — Century Solutions Group treats social engineering like the serious, evolving threat it is.

Because in cybersecurity, the human layer isn't a weakness you accept. It's a layer you protect.

If your business handles sensitive data, client relationships, or financial systems — and honestly, who doesn't — this is worth a conversation.

🔗https://centurygroup.net/

Data breaches are rising faster than ever in 2026, affecting individuals, businesses, and even governments. From weak email security to exposed IP address data, the risks are growing more complex and harder to detect. This blog breaks down the real reasons behind the surge, how attackers are evolving, and what you can do to stay protected in a connected world.

AI-driven malware exfiltrated credentials and documents via OpenClaw skills and malicious developer packages, while NuGet, npm, and Chrome extension attacks highlighted persistent supply-chain and client-side risks.