Trends in API security

Main article: Application Programming Interface (API)

2025: Trends in API safety - from the rise of leaks to the impact of AI

In recent years, the role has API become increasingly critical for organizations that prefer online to engage with customers and partners. According to one of the major service providers, CDN the company, Cloudflare already in 2022-2023 in their global network, more than 53% of the dynamic content HTTP in generated APIs.

The ongoing digital transformation and widespread use of Artificial Intelligence (AI) have cemented the API's leading role in the business processes of large organizations. The IDC study showed a sharp increase in the use of generative AI from 55% in 2023 to 75% in 2024.

According to the report, which TAdviser was shared by SQUAD at the end of June 2025, the downside of the technology is an increase in the number of incidents related leaks to APIs, as well as an increase in the number of known vulnerabilities in web applications with a built-in API. In 2025, the company Akamai published the results of a survey conducted among 800 + employees of companies in the Asia-Pacific region. These results showed that organizations regularly face API security incidents, and the average damage over the past 12 months ranges from 300 to 800 thousand. dollars

High-profile incidents and vulnerabilities in 2024

Let's start with leaks through API, TOP 3 in terms of the number of leaked records:

1. DELL - 49 million records^[1]

The hacker discovered the possibility of registering on the DELL partner portal, which does not require additional verification - just fill out the application form. Then, using the newly created account, the attacker, using a script, collected information about clients through the API using a seven-digit service tag. At the same time, the script worked for three weeks without any locks, generating up to 5,000 requests per minute.

^[2]

At the end of June 2024, hackers stole 33 million records related to Twilio's two-factor Authy authentication application. The leak included phone numbers, account IDs and some other non-personal data related to Authy users. Twilio later confirmed that the attackers were able to identify data related to Authy accounts, including phone numbers, due to the lack of authentication on one of the endpoints in the company's API.

3. Trello - 15 million^[3].

The Trello open API allowed private email addresses to be associated with service accounts. In January 2024, an unknown hacker tried to sell data from more than 15 million Trello accounts on the darknet, which he extracted using this API. Later, the hacker clarified the details of the attack. He produced a list of 500 million email addresses and forwarded them through an open API to determine if they were linked to Trello accounts. In the published database data contains more than 15115945 lines consisting of an email address, first/last name, login and link to an avatar.

In terms of vulnerabilities of the first place, the administrative portal Subaru is worthy. The researcher discovered a whole bunch of vulnerabilities that eventually allowed any user Internet to perform a number of critical operations related to the cars of this manufacturer:

• Remotely start, stop, block vehicles;
• Receive information about the current location of the car, as well as a complete history of locations over the past year;
• Receive personal data of customers: emergency contacts, addresses, payment information, car PIN, support call history, previous owners, odometer readings, sales history and much more.

Vulnerabilities worthy of special mention:

• The ability to reset the passwords of portal users without a confirmation token, that is, any Internet user could access the employee's account;
• Bypass two-factor authentication - the form could be disabled right in the browser;
• Authorization violations - any account had access to

multiple critical API endpoints (personal data, operations with a car,...).

Impact of AI on API services

The active use of AI in 2024 became the main growth driver for the API. Postman, which creates API building and testing tools, noted that their platform has 73% more AI-related traffic. At the same time, the use of services such as ChatGPT, Gemini, Mistral has grown tenfold.

Rapid implementation of AI services using API allowed companies to gain a competitive advantage and reduce their costs. At the same time, insufficient attention was paid to security issues that led to the publication of critical endpoints without adequate protection. Weak authentication or its absence, incorrect authorization, the absence of limits on the number of requests is just part of the examples.

In addition, AI agents often have access to customer information, which increases the likelihood of leakage. In many organizations have introduced prohibitions or significant restrictions on use of AI before elaboration of protection measures. This leads to growth "shadow" services in the IT infrastructure and also increases risks compromises and leaks. The use of AI in development allows you to speed up both code generation and deployment as a final product. At that at the same time, security commands cannot quickly detect and fix vulnerabilities. As a result, the number of tasks for correction is growing vulnerabilities, and some of them may be present for a long time in production environments.

Separately, it is worth paying attention to the vulnerabilities that go into set with popular tools and frameworks in the field of AI. Searching the database of known vulnerabilities cve.org shows multiple growth the number of vulnerabilities associated with AI:

• CVE-2022: 17
• CVE-2023: 68
• CVE-2024: 320

Examples of vulnerabilities include:

1.CVE-2024-3660 Keras code injection vulnerability·

Keras is a high-level API for the popular platform TensorFlow, which allows you to create and run models based on machine learning. Code injection vulnerability allowed attackers execute arbitrary code with the same permissions that the application has.

2.CVE-2023-43654 ShellTorch

TorchServe is a tool for maintenance and scaling PyTorch models. ShellTorch is a chain of 3 critical vulnerabilities in this tool, which led to remote execution code.

3.CVE-2024-0917 remote code execution in paddlepaddle

PaddlePaddle is a popular Chinese ML platform with, open source which is used in 430,000 companies in various sectors. IN 2024- m year baghunter demonstrated the possibility of remote code execution.

In 2025, OWASP published a new list of threats to large language models (Large Language Model^[4].

As in the previous version of the list, the threat remained in the first place, related to Prompt Injection, which occurs when user input "forces" the model to issue answers that it should not show. With sixth in second place was the threat associated with disclosure sensitive data. All this shows an increase in data breach risks for APIs using AI based on. LLM

Key API Protection Measures

Due to the growth of API use, the emergence of shadow services and by accelerating the publication of services, an inventory comes to the fore. Organizations must understand which endpoints and for whom they are published, what functions and parameters they are used for work. At the same time inventory on a periodic basis, such as using a scanner, is no longer enough - it must be done in real time. It associated with the widespread use of automated detection tools services and vulnerabilities on the Internet. In fact, they start attacking you on the first day after the API is published, and the first scans may affect service within a few minutes. For example, a popular service monitoring published resources Censys claims to be performing scan of the entire address space of the Internet at least once a day.

If the API inventory contains sensitive data information and on the presence or absence of authentication, you can quickly identify and Stop unwanted endpoint publications.

The second important factor in protection is the presence of an API description. One of the most popular standards is the Open API Specification (OAS). This description contains a list of methods, URLs, parameters, and codes responses that can be used in the API. Thanks to this description organizations can validate all compliance requests and responses and stop what is not in the specification, thus significantly reducing the attack surface.

The third boundary of protection should be limits on the number calls to endpoints. This allows you to stop attacks such as:

• Unlimited consumption of resources (for example, sending confirmation codes via SMS);
• Brute force passwords and/or user logins;
• Bulk data upload in the presence of an Insecure vulnerability

Direct Object Reference (IDOR), exploitation of this vulnerability as rule requires multiple requests with different object ID (for example: get/user/card/1, get /user/card/2, get /user/card/3, …) And finally, no one has canceled classic vulnerabilities such as SQL Injections, Code Injections, SSRF, XSS, XXE, Path Traversal, … Attacks, exploiting these types of vulnerabilities are well detected signature methods. The relevant rules provide the Web Application Firewall (WAF) or specialized API protection solutions.

Conclusions

The increase in use API leads to an increase in personal data leaks and vulnerabilities associated with this technology. Together with accelerated the introduction of digital services and AI organizations should not forget about the security of the APIs on which they are built. It is important to build a comprehensive a protection system that takes into account various threats and allows you to adequately react to them. The main protection functions shall be:

• Inventory
• Validation of Requests and Responses to the Open API Specification
• Restrictions on the number of cases per unit of time
• Ability to detect and block OWASP attacks

Just one information security incident, maybe cross out all efforts to develop API-based services, and in some cases may lead to the closure of the business. Comprehensive API protection will reduce cyber risks and ensure the stable development of digital business.

Notes