When AI Learns to Pick Locks: Security Now Covers Permissions, Supply Chains and Recovery
AI is reshaping attack and defense at once. Security now means controlling what AI can access, vetting the software supply chain and being able to recover after a breach.
Cybersecurity in the AI era goes beyond installing antivirus software. It is an industry that manages what people and AI can access, what actions they take, and how a company recovers when something goes wrong.
Think of the internet as a huge city. Websites are buildings, passwords are keys, and the common software used to build programs is the same brand of lock fitted to many buildings. Now a tool has arrived that can find weaknesses in those locks quickly: AI. For defenders it is a precision inspection kit. For attackers it can be a powerful tool too.
Alarm bells across the internet city
The incidents discussed here were disclosed around October 21, 2026. But a vulnerability discovery, a malicious software release and a breach of certificate trust are different events. Lumping them together as one big AI-driven hack would be inaccurate.
| Incident | Plain-language analogy | What is at risk |
|---|---|---|
| PHP vulnerability | A flaw found in a common lock fitted to many buildings | Under certain conditions, an unauthenticated attacker could run commands on a server, which required an urgent fix. |
| Node.js vulnerabilities | Several flaws found in shared building equipment | Ten issues were fixed in total. Four were rated high severity, including remote code execution and a bypass of server identity checks. |
| LiteLLM malicious release | Parts from the official warehouse arrived with a bug planted inside | Certain releases of a tool that connects AI services contained code to steal passwords and access keys and to run remote commands. |
| eMudhra certificate problem | The stamp of an ID-issuing agency can be forged by others | Weak private keys in eight intermediate certificates created a risk that fake certificates could be made to impersonate legitimate servers. |
One distinction matters here. "The door has a flaw" and "a thief has already come in" are not the same statement. Some government websites that went down because of the PHP issue were taken offline as a precaution to block intrusion.
The eMudhra case is also better understood as a loss of certificate trust because the keys themselves were weak, not simply as someone stealing a strong secret key. Browsers, operating systems and related software must stop trusting those certificates, which widens the scope of the response.
Three ways AI changes the picture
Finding weaknesses gets faster
Skilled specialists once had to study code and systems for a long time. AI can assist with and automate part of that work. In the PHP case, AI-assisted research helped show that an old problem could actually be exploited.
The catch is that attackers can use the same capability. The UK's National Cyber Security Centre has assessed that AI will make some stages of cyber intrusion more effective and efficient, and will further shorten the time between a vulnerability's disclosure and its exploitation.
An analogy: a burglar used to rattle doors house by house. Now he has an assistant that can check more doors, faster. That does not mean AI can open every door.
AI needs a badge too
A chatbot that only answers questions is different from an AI agent that carries out real work. If an agent reads company files, calls programs and modifies data, it needs a unique identity and limited permissions, just as a person does.
What if an "AI employee that sorts receipts" were handed a master badge that opens the company's accounts, HR records and customer files? It looks convenient, but if that AI is attacked or misbehaves, the damage grows with it. So each AI should be allowed only the data and functions it needs, its actions should be logged, and its permissions should be revocable at once.
Going forward, the key security question is not only "who logged in" but "which AI did what, with whose permission."
Legitimate tools need scrutiny too
The LiteLLM incident shows that users can be hurt even without hunting for suspicious hacking tools. It was a software supply chain attack, in which a particular version on the official distribution path was contaminated.
Put simply, the supplier of ingredients was attacked, not the restaurant. However clean the restaurant is, contaminated ingredients cause problems. Security therefore has to cover not just finished services but developer accounts, outside components and the release process.
Security is not one product
The easiest way to understand AI security is to split defense into three layers. This framework is for understanding the industry. It does not mean any single product stops every incident.
- Keep intruders out: Verify the identity of users and AI, and allow access only to the data they need. "Zero trust," which does not trust anyone just because they are inside the company, belongs here.
- Catch them fast if they get in: Detect abnormal behavior on computers, servers and in the cloud, block attacks and investigate.
- Get back up even if breached: Keep backups isolated so attackers cannot easily alter or delete them, and restore operations from clean data.
Seen this way, security is not about hiring one guard. It is closer to running badge management, a monitoring room and emergency recovery facilities together.
Why this industry matters
To understand security demand, it helps to look beyond "there are a lot of hacking headlines" and ask what companies must solve first to connect AI to real work. Giving AI data and the power to act requires access controls, activity logs and ways to revoke permissions.
Defenders use AI as well. Security products are evolving to use AI and automation to link large numbers of alerts and assist investigation and response. In other words, a market for using AI to protect and a market for protecting AI itself are forming together.
For investors, recurring service usage also matters. CrowdStrike, for example, reports subscription revenue separately. But being subscription-based does not guarantee growth. Security companies are also exposed to falling renewal rates, price competition and technological change.
US-listed security stocks
These companies are all "security stocks," but their starting points and strengths differ. The analogies are meant to aid understanding, and in practice their businesses overlap and are expanding.
| Company, Ticker | In plain terms | Core role | Link to the industry |
|---|---|---|---|
| Palo Alto Networks, PANW | A company's integrated security HQ | A platform spanning network, cloud and security operations. | Demand to consolidate and manage multiple security functions |
| CrowdStrike, CRWD | Digital guards for computers and servers | Starts with endpoint protection and extends monitoring to identity, cloud and AI. | Detecting and responding to abnormal behavior after intrusion |
| Fortinet, FTNT | Security checkpoint for offices and factories | Firewalls, network security and equipment for industrial sites. | Protecting real networks and operational and industrial systems |
| Zscaler, ZS | Access control that opens only the rooms you need | Connects users to permitted applications, not the whole corporate network. | Zero-trust access control in cloud environments |
| Okta, OKTA | Badge office for people and AI | Identity and access management, and identity control for AI agents. | Demand to give each AI an owner, permissions and a way to stop it |
| Cloudflare, NET | A shield in front of websites | Defense against web attacks, DDoS, malicious bots and API abuse. | Protecting public-facing services and connection paths |
| Rubrik, RBRK | Data vault and recovery center | Immutable backups and data recovery after an attack. | Securing the ability to resume operations, not just "perfect prevention" |
For those studying the sector, a useful start is to sort by role: PANW as the integrated platform, CRWD as detection and response, OKTA as identity, RBRK as recovery. This is a study classification, not a ranking of returns or an order of purchase.
Also, these companies do not directly fix every vulnerability described here. Root flaws in common software such as PHP and Node.js require fixes from those projects and adoption by the organizations that use them. Security products cannot replace updates themselves.
Approaching the sector through ETFs
For those who find it hard to pick company-level winners, ETFs that hold many security companies are one option. But these funds are listed on US exchanges, which does not mean they hold only US companies.
| ETF | Annual expense ratio | Portfolio character | What to compare |
|---|---|---|---|
| CIBR · First Trust Nasdaq Cybersecurity ETF | 0.58% | Core security companies plus some related technology and service firms. | Which large tech companies sit alongside pure security names |
| HACK · Amplify Cybersecurity ETF | 0.60% | Holds security hardware, software and services together. | Exposure to adjacent businesses such as networking equipment, semiconductors and defense |
| BUG · Global X Cybersecurity ETF | 0.50% | Holdings must meet a minimum 50% cybersecurity revenue threshold. | A structure relatively concentrated in security specialists |
| IHAK · iShares Cybersecurity and Tech ETF | 0.47% | Invests in security hardware, software and services in developed and emerging markets. | Country mix and holdings differences from other ETFs |
Expense ratios are based on confirmed fund-sponsor disclosures and may change. CIBR's page shows costs as of February 2, 2026. HACK and BUG show product information as of October 21, and IHAK shows costs from its current prospectus. The figures do not include trading commissions, bid-ask spreads, currency conversion costs or taxes.
Actual holdings also differ. CIBR's top holdings on October 19 included Fortinet, CrowdStrike, Cisco, Palo Alto Networks and Broadcom, while BUG's top holdings on October 21 included Okta, CrowdStrike, Fortinet, Palo Alto Networks and Zscaler. That is why funds carrying the same industry label can perform differently.
One way to sort them: CIBR as a starting point for surveying the whole industry, BUG as a comparison when the weight of security specialists matters, HACK as a comparison that includes adjacent businesses, and IHAK as a candidate when cost and global exposure are weighed together. This is a classification for understanding fund structure.
A growing industry is not the same as a good investment
More hacking does not mean every security stock rises. Even as an industry grows, competition and price cuts can erode profitability, and investors may already have priced in high growth.
A security company's own operational failures are a risk too. In 2024, a faulty content update from CrowdStrike crashed some Windows systems. It was not a hack, but it shows that product reliability and release quality matter when evaluating security companies.
When analyzing a company, these questions are worth asking:
- Are customers still growing? Look at the trend in new contracts and customer expansion rather than short-term events.
- Do existing customers spend more? Check whether renewals and add-on product sales continue.
- Does it actually generate cash? Look at operating cash flow and free cash flow, not just revenue.
- Is shareholder value growing too? Check dilution from stock-based compensation and new share issuance.
- Is a good future being bought at too high a price? Compare growth and profitability with the current valuation.
An ETF spreads single-company risk, but it is not insurance against a decline in the whole industry. It can be affected by technology change, competition and market declines.
Security habits to apply today
- Do not put off updates: A security fix protects you only once it is actually applied to your devices, browsers and apps.
- Do not reuse passwords: Use a different password for each account, and use a password manager.
- Turn on multifactor authentication: Start with email, financial and cloud accounts.
- Back up important data separately: Keep backups isolated so an attacker cannot damage the original and the backup together, and test your recovery.
- Do not give AI a master key: Allow only the data and functions it needs, and require separate approval for key actions such as deletion, payment and deployment.
Website and app operators should not assume "we removed the bad version, so we're done" once contamination of an outside tool is confirmed. An environment that installed or ran an affected LiteLLM release should be treated as compromised. Operators should consider rebuilding it clean and rotating any access keys and passwords that may have been exposed.
This is an educational explanation to help readers understand the industry and its products. It does not assess whether current share prices are appropriate or offer individual buy or sell advice.
Insight Times Editorial Desk
