In August 2025, Apple and WhatsApp released emergency advisories for CVE-2025-43300 and CVE-2025-55177, a pair of vulnerabilities that, chained together, had already been used in the wild to compromise the mobile devices of fewer than 200 users over the preceding 90 days [1, 2]. Independent reporting identified many of the targets as journalists, human-rights defenders, and civil-society figures [3]. The chain required no clicks, no downloads, and no error on the victim’s part: receiving a WhatsApp message was sufficient. Both CVEs were added within weeks to the U.S. Cybersecurity and Infrastructure Security Agency (CISA) Known Exploited Vulnerabilities (KEV) catalog [4], mandating remediation across federal networks.
Zero-click chains of this shape have become the defining offensive capability of commercial spyware. Unlike phishing, which relies on social engineering, and unlike drive-by compromise, which relies on the victim visiting an attacker-controlled URL, a zero-click chain weaponizes the infrastructure that modern mobile operating systems use to render inbound media. Knowledge of the victim’s phone number is sufficient; no further action on the victim’s part is required. The asymmetry of this attack model (bounded defender effort against unbounded attacker opportunity) makes zero-click delivery the most consequential primitive in targeted surveillance today [5].
What makes the 2025 chain technically striking is its genealogy. Both vulnerabilities fire in code written in C, a language first standardized in 1989 whose memory model (Section II) provides no enforcement of bounds, type safety, or lifetimes. The exploitation stage is a heap out-of-bounds write in an image parser, a bug class whose stack-based ancestor was first formalized by Aleph One in 1996 [6] and whose heap-resident variant has ranked second in the MITRE CWE Top 25 in 2024 (score 45.20, 18 entries in the CISA Known Exploited Vulnerabilities catalog) [7, 8] and remained the highest-ranked memory-corruption class in 2025 [9]. Industry-wide, memory-safety bugs account for approximately 70 % of all CVEs patched annually by major software vendors - a figure confirmed independently by Microsoft’s Security Response Center across twelve years of patch analysis [10] and by Google’s Chromium team across hundreds of high-severity browser bugs [11]. Despite repeated calls from CISA and the U.S. Office of the National Cyber Director for migration to memory-safe languages [12], the media-parsing frameworks at the core of every major mobile platform remain overwhelmingly C and C++. The result is a structural mismatch: a decades-old memory model is entrusted with the first parsing step of every image delivered to every networked device.
This is not the first time the pattern has produced high-profile civilian casualties. In 2021, Citizen Lab and Google Project Zero reverse-engineered FORCEDENTRY (CVE-2021-30860), an NSO Group Pegasus zero-click chain with nearly identical architecture: iMessage delivery, CoreGraphics image parsing, heap corruption [13, 14]. Four years later, the 2025 WhatsApp–ImageIO chain reproduces the same three-step pattern in a different ecosystem, messenger, image format, and parser, with the same outcome. This cross-platform recurrence is the central empirical observation motivating this study. As this manuscript was being finalized, two further write-primitive Linux kernel root exploits were disclosed in quick succession: Copy Fail (CVE-2026-31431, April 2026, AF_ALG crypto subsystem) [15] and DirtyFrag (CVE-2026-43284, May 2026, xfrm-ESP networking) [16] - providing real-time evidence that the broader write-primitive exploit class documented in Section II-F is ongoing and not confined to the iOS ecosystem.
Research question. This paper asks: how do decades-old memory-corruption patterns in C enable modern zero-click exploit chains, and what does the 2025 WhatsApp–ImageIO attack reveal about the persistence of these vulnerabilities in industry-critical software?
Contributions. This paper makes three contributions:
A reconstruction of the 2025 zero-click chain from public primary sources (vendor advisories, patch diffs, and independent reverse engineering [17]), covering the delivery primitive (CVE-2025-55177, CWE-863) and the exploitation primitive (CVE-2025-43300, CWE-787), and showing how the two stages compose into a complete, interaction-free compromise.
A pedagogical bridge from the Exploit Education Phoenix heap-two exercise to CVE-2025-43300, demonstrating that the core mechanic (a heap buffer overflowed into adjacent function-pointer-bearing data) is the same primitive at two drastically different complexity levels. The bridge is intended for readers whose prior exposure to memory corruption stops at textbook stack smashing.
A structural argument that the FORCEDENTRY → WhatsApp–ImageIO lineage, spanning four years and two independent ecosystems, is evidence that the current mitigation stack (ASLR, sandboxing, pointer authentication) has not closed the fundamental attack class; it only raised its price.
Scope and non-goals. This paper does not present a new exploit, a new vulnerability, or proprietary reverse engineering. All technical material is drawn from publicly disclosed sources. The Phoenix heap-two exercise is treated as a controlled pedagogical artifact, not a substitute for the real chain.
Roadmap. Section II introduces the C memory model, the two CWE classes at play, the iOS media pipeline, the FORCEDENTRY precedent, and the decade-long recurrence of write-primitive exploits across iOS and Linux. Section III describes our CVE selection criteria, analysis framework, and Phoenix setup. Section IV is the core technical analysis of the 2025 chain. Section V reflects on why the current mitigation stack fails to close this attack class. Section VI concludes.