Loading story…
Microsoft is expanding agentic software development across GitHub Copilot, Aspire, Windows development environments, WSL, and model-context-protocol integrations, with Distinguished Engineer David Fowler saying manual line-by-line coding is becoming less central. However, automated code generation continues to produce material security defects: Veracode’s 2026 benchmark found known vulnerabilities in roughly 44% of AI code-generation tasks, while Sonar’s evaluation of GPT-5.6 variants found higher bug and vulnerability densities than GPT-5.5 despite improved functional-task pass rates, including cryptographic misconfiguration, insecure resource handling, and concurrency flaws. Microsoft is deploying its MDASH agentic scanner across engineering environments to identify vulnerabilities at scale.
Security researchers also found exploitable weaknesses in rapidly built “vibe-coded” applications, including client-side-only authentication, browser-exposed API keys, permissive or missing Supabase Row-Level Security policies, and internet-exposed internal tools. Tests of GitHub Copilot’s GPT-4o recommendations showed proposed fixes for a Node.js path-traversal flaw could be bypassed by relative or URL-encoded traversal payloads. Organizations should require human architectural and security review of AI-produced changes and retain secure-by-default libraries, SAST, software-composition analysis, secret scanning, pull-request controls, backend secret management, and deny-by-default database access policies.
Track how attackers are adapting to this technology.
10 events from the most recent confirmed update back to the earliest known activity.
Microsoft's Aspire team stated that AI agents can effectively write code, but generating code is distinct from delivering a complete working application. Aspire supports agent actions including starting services, inspecting logs and telemetry, restarting failed components, and retesting applications.
GitHub added Windows development environments to its Copilot coding agent, enabling the agent to build and test Windows-targeted projects, run linters, and verify builds.
Veracode's 2026 GenAI Code Security Report found that approximately 44% of tested AI code-generation tasks produced code containing a known vulnerability in a controlled benchmark.
An evaluation of GPT-4o through GitHub Copilot found that its proposed checks for a deliberately vulnerable Node.js bank-statement API could be bypassed by relative and URL-encoded traversal payloads. The assessment concluded that LLM suggestions lack sufficient source-to-sink context to serve as standalone vulnerability remediation.
Microsoft Distinguished Engineer David Fowler said that “typing code is absolutely over,” describing a shift toward AI agents performing more implementation work while developers retain responsibility for architecture, validation, testing, dependencies, and security review.
Microsoft deployed its MDASH agentic vulnerability-scanning system across major engineering environments, including Windows, Azure, and identity systems. The Windows security team said it enabled deeper vulnerability hunting involving the Windows kernel, Hyper-V, and networking stack.
Sonar evaluated GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.5 on 4,444 Java tasks using SonarQube analysis. Sol achieved the highest functional pass rate, but both GPT-5.6 variants had higher bug and vulnerability densities, particularly for concurrency defects, cryptographic misconfiguration, and insecure resource handling.
After Wiz shared its findings with Lovable's security team, Lovable published security best-practice guidance and collaborated on a global system-prompt change intended to discourage embedding secrets in client-side code.
Wiz Research identified client-side-only authentication, browser-exposed credentials, permissive or absent Supabase Row-Level Security, and publicly accessible internal applications in vibe-coded apps. Reported examples included an enterprise game exposing users' PII and IP addresses.
Wiz released baseline open-source security rules files and the prompt used to generate them, targeting common languages and frameworks for use with AI coding assistants. The guidance promotes concise, scoped rules and conventional controls such as SAST, SCA, secret scanning, and pull-request review.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
5 references tracked. Mallory keeps watching after this page renders.
windowslatest.com
Open sourcesonarsource.com
Open sourcewiz.io
Open sourcewiz.io
Open sourcenodejs-security.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Multiple studies and industry reports found that **AI-generated or "vibe-coded" software is frequently shipped with serious security flaws**, including XSS, SQL injection, hardcoded secrets, unsafe `eval` usage, weak authentication logic, wildcard CORS, exposed `.env` data, and insecure dependencies. One analysis of 5,600 publicly accessible AI-built applications reported more than 2,000 high-risk vulnerabilities and over 400 exposed secrets, while a separate review of 549 self-described AI-generated repositories found widespread committed secrets, injection issues, hardcoded passwords, and other common sanitization failures. Veracode and other researchers similarly warned that code produced by large language models often maps to **OWASP Top 10** weakness classes and that repeated AI-assisted edits, poor prompts, and excessive agent permissions can further increase risk. At the same time, researchers and vendors highlighted both the promise and fragility of AI-assisted security tooling. Trail of Bits reported that OpenAI Codex goal-based prompting helped uncover previously unknown bugs in projects including **Rust**, **curl**, **zlib**, and **Keycloak**, but stressed that human experts were still needed to scope goals, validate findings, and handle disclosure. New academic work also showed that **LLM-based vulnerability detectors can be easily misled**: the ALIBI framework used adversarial code comments to evade four detectors with success rates above 90%. In response to growing concern over agent reliability, Amazon backed the Lean language project to strengthen formal verification for AI agents, while security experts argued that theorem proving can improve assurance for high-risk components but should be applied selectively as part of broader defense-in-depth.
Aug 4, 2026
The widespread adoption of generative AI (GenAI) tools in software development has significantly increased productivity, enabling developers to document, write, and optimize code at unprecedented speeds. According to a 2023 McKinsey study, organizations have rapidly integrated AI into their development workflows, with 83% using AI for code creation and 57% relying on AI-powered coding tools as a standard practice. However, this surge in AI-assisted coding has introduced new security risks, as traditional security models focused on perimeter or infrastructure controls do not adequately protect the data and code generated by these tools. Studies have revealed that nearly half of code snippets produced by popular AI models contain vulnerabilities, underscoring the prevalence of insecure code generation. High-profile incidents, such as Samsung's 2023 ban on ChatGPT following a sensitive code leak, highlight the real-world consequences of insufficient safeguards when using GenAI in development environments. The responsibility for securing data and code remains with developers, even as cloud providers secure the underlying infrastructure. The rapid pace of AI-generated code has outstripped the ability of traditional secure coding training to keep up, shifting the focus from training human programmers to ensuring that AI systems themselves are capable of secure coding. Industry experts note that AI is currently less effective at producing secure code than human programmers, with multiple studies and reports from sources like Schneier on Security, Veracode, and SC Media confirming this trend. The volume of vulnerabilities continues to rise, with over 47,000 publicly known vulnerabilities expected in a single year and at least 130 new vulnerabilities reported daily. This ongoing wave of vulnerabilities leads to constant exploitation and patching, further emphasizing the need for secure coding practices at the AI level. While AI has delivered substantial productivity gains—developers report 30% to 40% increases—these benefits are undermined by the security shortcomings of AI-generated code. The industry is now at a crossroads, where the imperative is to teach AI systems to code securely, rather than relying solely on human oversight or post-development security reviews. Integrating security into the AI coding process and providing developers with tools that embed data protection are seen as essential steps to address these emerging challenges. The shift towards AI-driven development necessitates a reevaluation of security strategies, focusing on proactive measures that align with the realities of modern software engineering. Without such changes, organizations risk exposing themselves to significant security threats stemming from the very tools designed to enhance their productivity.
Jun 29, 2026
The rapid adoption of AI-powered code generation tools such as GitHub Copilot, ChatGPT, and Amazon CodeWhisperer has fundamentally changed the software development landscape, introducing new security challenges for enterprise application security teams. Unlike traditional human-written code, AI-generated code often lacks clear provenance, making it difficult to verify its origin or ensure compliance with organizational security policies. This shift has led to the emergence of 'shadow code'—machine-generated code that may bypass standard security reviews and evade detection by traditional static and dynamic analysis tools, increasing the risk of invisible vulnerabilities in production systems. Generative AI models can introduce unique threats, including the creation of 'hallucinated' packages—references to non-existent or malicious libraries that may be inadvertently included in enterprise applications. Additionally, the language-based nature of large language models (LLMs) opens new attack surfaces, such as prompt injection and jailbreaking, where malicious inputs can manipulate model behavior or bypass safety constraints. As organizations accelerate the integration of AI into development workflows, application security programs must adapt to address these novel risks and build trust in the security of AI-powered software.
Jun 29, 2026