Saturday, November 21, 2026

Top 5 This Week

Related Posts

SusVibes: 11.8% of AI-generated solutions are secure | Brief AI


In the SusVibes evaluation, programmers deprived of security guidelines produced secure solutions in only 11.8% of cases. However, their responses could be functional up to 57% of the time. For organizations, controls concern imported components as well as human responsibilities and applications created outside of IT teams.

Budgets poorly follow software development

Security budgets generally grow more slowly than software development, said Darren Meyer, research engineer at Endor Labs. According to him, the adoption of AI widens this gap. Application security is often insufficiently funded, while assisted programming accelerates development. Some decision-makers may still view the protection of AI systems as optional.

Darren Meyer believes that both quality tools and specialists in application and product security are necessary. They must be able to collaborate with developers of various profiles and understand the technical, compliance and security aspects of AI. The volume to be reviewed may exceed the capabilities of a fully manual review. Leslie Nielson, information systems security manager at information security vendor Mimecast, warns that a reviewer saturated with low-risk changes ends up committing without proper review.

Darren Meyer also describes organizations divided between very strong support for AI and a marked refusal. These divisions can complicate the tasks of security managers, including in the trades. Insufficiently explained restrictions risk being seen as an obstacle to innovation. Thomas Segura, senior technical editor at GitGuardian, advocates a strategy that combines productivity with security requirements.

Imported components can escape controls

A dependency missing from the manifest, called a “ghost”, may not be detected by a scanner and may contain a vulnerability. Third-party libraries are also a point of vigilance. Software composition analysis and supply chain tools can help identify the components used, their flaws, and associated legal or compliance challenges.

Alex Ștefănescu, security expert and technology activist, warns about packages invented by agents or completion functions, particularly in GitHub Copilot. A statistically plausible continuation after an import statement does not guarantee that a package exists on PyPI, this person explains. According to Alex Ștefănescu, attackers register suggested names on npm or PyPI and add functions that make the packages appear credible. Importing them into an application can cause damage.

Machine learning models and datasets uploaded to Hugging Face or other platforms also pose risks, regardless of code generation. According to Alex Ștefănescu, a backdoor can remain despite checks on both ends, activate during model integration and cause a leak of company information. The use of programming wizards can also result in sensitive data being unintentionally inserted into code bases.

Security managers can define procedures for downloading and integrating these external resources, then include them, where relevant, in agent instructions. These procedures may include automatic scanning for malware or backdoors. They can also limit the authorized models to those of verified publishers or impose internal tests in an isolated environment.

Mimecast reserves approval for people

At Mimecast, each change delivered with AI assistance is the responsibility of a person identified in the integration request, says Leslie Nielson. Validation belongs exclusively to humans. He recommends combining automatic checks with human judgment: his company automates repetitive checks on known dependencies, secrets and vulnerability patterns. Access, authentication and sensitive data nevertheless systematically pass before a specialist, even in the absence of automatic alert.

The evening letterFree · 7 p.m.

Applied AI, concretely?

Use cases that really work, deciphered every evening in 5 minutes. Free.

Included upon registration: our selection of the best AI guides & comparisons.

Choose your pace

Free · No spam · Unsubscribe in 1 click

Bianca Lankford, vice president of engineering at Datadog, recommends not pre-trusting generated code. Security teams are already bringing static analysis, secret searching, and dependency checks closer to each integration request, and even the publisher, to the developer. According to her, companies define authorized assistants, restrict their access and teach reviewers to spot obsolete components, poorly validated inputs and overly permissive settings.

For Bianca Lankford, the use of a specialist must depend on the risk, not on the origin of the code. She cites authentication, cryptography, payments, personal data and infrastructure, with automatic guidance possible through accountability rules on code. Other signals include the extent of modifications, the addition of dependencies, and changes in permissions or network exposure.

Alex Ștefănescu, for his part, distinguishes between software production and understanding how it works. According to this expert, maintained documentation, robust testing and keeping important architectural decisions in writing can save time during a security incident or complex diagnosis. These resources should be produced and maintained by humans for other humans, rather than entrusted to automatic generation.

Business teams develop their own applications

IT departments are increasingly calling on business users to create their applications, while Claude Code, Cursor, GitHub Copilot and Amazon Q Developer are among the main assistants used by developers. Darren Meyer is seeing significant adoption among data analysts, marketing teams, and researchers. These teams, traditionally unfamiliar with developing their own software, are manufacturing more small tools integrating libraries and AI models, without always knowing the risks.

An IT manager recounts on Reddit his discovery of the work of a young graduate practicing “vibe coding”, several months after the latter’s recruitment. According to his testimony, the young graduate had designed, in a domain over which he had control, a web application using customer data and private contracts. Supabase’s free tier was used to store information, while its GitHub repository directly powered deployment on Cloudflare’s serverless infrastructure.

The manager claims that the employee worked directly in production, with servers distributed around the world, while the company was in the European Union. Asked about the examination of the code and the precautions taken, the employee only cited the private nature of the GitHub repository. The CEO then reportedly dismissed the IT manager’s concerns.

Darren Meyer believes that this development outside the usual circuits, associated with limited knowledge of application security, promotes risks. Security leaders can reach out to affected teams early and consider tailored training for analysts, marketers, and researchers on the dangers of AI tools and libraries.

SusVibes tests functional code security

SusVibes is based on 186 real feature requests from open-source projects. Danqing Wang of the Language Technology Institute at Carnegie Mellon University, who participated in this evaluation, explains that the failures observed were not simple negligence. Detecting them required considering the possibilities of an attacker, beyond the expected operation.

Katie Paxton-Fear, security specialist at Semgrep, believes that the improvement of wizards in producing working programs encourages developers to place more trust in them and reduce their vigilance. The volume of code produced contributes to security managers’ concerns. AI knows how to look for defects in software, but it also introduces its own errors when it writes code.

Shredder Smith
Shredder Smith
Shredder Smith is the lead curator and digital persona behind topaitools4you.com, an AI directory dedicated to "shredding" through industry hype to identify high-utility software for everyday users. Smith positions himself as a blunt, no-nonsense reviewer who vets thousands of emerging applications to filter out overpriced "wrappers" in favor of tools that offer genuine ROI and practical productivity. The site serves as a watchdog for the AI gold rush, providing categorized rankings and transparent reviews designed to help small businesses and creators navigate the crowded tech landscape without wasting money on low-value tools.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Popular Articles