By rssfeeds-admin 
The catalyst for Apex is a structural breakdown in how software security is being practiced. AI coding agents are generating and merging code at machine scale Stripe’s coding agents alone merge 1,300 pull requests per week, while some engineering teams spend over $1,000 daily in AI tokens per engineer with zero human code review.
Traditional scanners and human-led assessments cannot keep pace with this velocity. Apex was built as the adversarial verification layer: a separate agent that attacks the running application exactly as a real attacker would, catching vulnerabilities before they become breaches.
Apex operates across three deployment modes. In CI pipelines, it validates every deploy against a sandboxed replica of the application, mapping the attack surface and attempting exploitation before code merges.
Against production, it continuously surfaces exploitable weaknesses in real time. It also supports on-demand testing against any target replacing the quarterly PDF engagement with a feedback loop that operates at the speed of modern threats.
To validate its capabilities, PensarAI built Argus, an open-source benchmark of 60 self-contained, Dockerized vulnerable web applications purpose-built for evaluating offensive security agents.
Existing benchmarks were deemed insufficient: the most widely used suite, XBOW’s 104-challenge set, is 70% PHP, covers single-vulnerability targets, and lacks GraphQL, JWT algorithm confusion, race conditions, prototype pollution chains, WAF bypass, and multi-tenant isolation scenarios.
Argus spans the frameworks dominating production: Node.js/Express (40%), Python/Flask/Django (20%), multi-service architectures (25%), Go, Java/Spring Boot, and PHP.
It introduces categories no other benchmark covers: WAF and IDS evasion, multi-step exploit chains requiring up to 7 chained vulnerabilities, multi-tenant isolation failures, race conditions and business logic flaws, modern authentication bypasses (JWT, OAuth, SAML, MFA), and cloud/Kubernetes infrastructure attacks. Difficulty is calibrated across 2 easy, 27 medium, and 31 hard challenges.
271 Vulnerabilities Across 60 Applications
Apex was pointed at all 60 Argus challenges in full black-box mode using Claude Haiku 4.5, the smallest, cheapest model available, to isolate architectural gains over raw model capability.
Apex achieved a 35% pass rate, outperforming PentestGPT (30%) and Raptor (27%). On the top 10 hardest challenges using Claude Opus 4.6, the gap widened substantially: Apex solved 80%, PentestGPT reached 70%, and Raptor hit 60%.
Across the full run, Apex discovered 271 unique vulnerabilities spanning SQL injection, SSRF, NoSQL injection, prototype pollution, SSTI, XXE, race conditions, IDOR, auth bypass, CORS misconfigurations, command injection, and path traversal. The average cost per challenge was approximately $8, with the entire 60-challenge run on Haiku costing under $500.
Notable solves included a 7-step race-condition double-spend in a fintech transfer endpoint, a multi-tenant SSRF chain pivoting through a shared cache to extract API keys from neighboring tenants, and SpEL injection to RCE a Java Spring Boot application — all in under 15 minutes.
Apex’s documented failure modes are instructive. Last-mile execution, completing the final credential extraction step after a successful SSRF chain, emerged as the dominant gap. Decoy flags misled the agent twice, and complex multi-step chains such as CI/CD pipeline poisoning and Kubernetes compromise exceeded the 30-minute budget.
Both Apex and the Argus benchmark are available as open source on GitHub today.
Follow us on Google News, LinkedIn, and X for daily cybersecurity updates. Contact us to feature your stories.
The post Apex – AI-Powered Pentester Attacks Apps in Black-Box Mode to Find Vulnerabilities appeared first on Cyber Security News.
Discover more from RSS Feeds Cloud
Subscribe to get the latest posts sent to your email.