The incidents covered in today's briefing highlight a recurring pattern: attackers achieving significant impact through trusted access rather than technical exploitation. A ransomware attack reached Coca-Cola's Fairlife subsidiary and halted US production, demonstrating that organisational scale and resources do not eliminate the possibility of operational disruption once attackers establish a foothold.

Two other incidents illustrate how limited initial access can create broader exposure. A compromised npm package in the AsyncAPI ecosystem and a single employee's stolen credentials at Suno converted a single point of access into broader compromise — one through a software dependency, the other through a developer account. The security challenge is therefore not only determining whether access is trusted, but understanding what that access enables if compromised.

Social engineering follows the same pattern. ClickFix-style lures appeared across several unrelated campaigns today — ACR Stealer, TELEPUZ, and ClickLock — relying on users to execute attacker-controlled commands. The sentencing of two Scattered Spider members over the £29 million TfL attack further demonstrates that exploitation of identity and trust relationships remains a significant driver of cyber incidents.

Coca-Cola disclosed in a Form 8-K filing that its Fairlife dairy subsidiary detected unauthorised third-party access to its systems, including production-related systems, connected to a ransomware attack. The company activated incident response protocols, notified law enforcement, and confirmed production at Fairlife's US facilities has been temporarily suspended, while Canadian operations remain unaffected. No ransomware group has claimed responsibility, and Coca-Cola has not disclosed whether data was stolen or an extortion demand received.

Group-IB discovered ClickLock Stealer, a previously undocumented modular macOS malware uploaded to VirusTotal on 9 June 2026 with zero detections at analysis time. The malware is likely distributed via ClickFix phishing pages, uses compromised WordPress domains and Telegram infrastructure, and kills system processes to force victims into entering their password. It targets data from eight browsers, 31 crypto wallet extensions, seven password manager extensions and macOS Keychain, and has affected at least 100 victims across 33 countries since May 2026.

Symantec identified Spirals, a previously unseen Rust-based ransomware deployed against a South Asian IT services company in June 2026 via a double extortion attack. Attackers gained initial access through a compromised IIS web server and an ASP.NET web shell, then used PsExec and WMI to spread a payload disguised as bitsadmin.exe across the network within 24 hours. The malware encrypts files using per-file AES-128 keys wrapped with an ECDH P-256 public key and threatens data leaks after six days via a Tor negotiation portal.

Thalha Jubair, 20, and Owen Flowers, 18, were each sentenced to five years and six months at Woolwich Crown Court for the 2024 cyberattack on Transport for London under the Computer Misuse Act. The intrusion cost TfL £29 million in recovery expenses, disrupted 148 internal systems, and forced 27,000 employees to attend offices for in-person password resets. The National Crime Agency identified both men as leading members of Scattered Spider, a network linked to attacks using social engineering, SIM-swapping and credential theft against airlines, retailers and insurers.

OpenAI developed GPT-Red, an automated red-teaming model trained via self-play reinforcement learning to discover prompt injection vulnerabilities and adversarially train production models. The model achieved an 84% attack success rate against GPT-5.1 in an indirect prompt injection benchmark, compared with 13% for human red-teamers, and compromised a live AI-powered vending machine agent by altering prices and cancelling orders. GPT-5.6 Sol, trained using GPT-Red's attacks, now fails on only 0.05% of GPT-Red's direct prompt injection attempts, a sixfold improvement over the prior production model.

Researcher wunderwuzzi demonstrated that ANSI escape codes rendered by macOS Terminal could trigger DNS requests, enabling data exfiltration when LLM-integrated CLI tools were hijacked via indirect prompt injection. In a proof-of-concept, a malicious spreadsheet cell instructed an AI agent to encode stolen customer names into an escape sequence, which Terminal interpreted and leaked via DNS. Apple was notified in December 2024 and addressed the issue in macOS Tahoe 26.1, released on 3 November 2025, after which the escape sequence no longer triggers DNS requests.

Microsoft identified a supply chain compromise of five @asyncapi npm package versions, published on 14 July 2026 after attackers stole a privileged bot token through a vulnerable GitHub Actions workflow. The injected loader executes at import time rather than through install scripts, spawning a hidden process that fetches an encrypted Miasma runtime via IPFS for command-and-control and persistence. Credential-harvesting and propagation modules were implemented but disabled, while active C2 infrastructure operated from IP address 85[.]137[.]53[.]71 across three ports.

Elastic Security Labs identified TELEPUZ, a modular Windows malware family spreading through CLICKFIX-VIDAR chains that deliver VIDAR variants, stagers and payload components. The malware uses indirect syscalls, UAC bypass techniques and Windows services for persistence while communicating with command servers through WebSockets and encrypted channels. Elastic observed TELEPUZ modules providing keylogging, browser data theft, web injection, screenshot capture and command execution capabilities across analysed samples.

Kaspersky identified GoSerpent, a Go-based remote access Trojan used since late 2025 against government and diplomatic entities in Southeast Asia, with encrypted command-line arguments and ChaCha20-protected command-and-control communications. The malware deployed ThumbcacheService to collect documents, archive files into thumbcache_605a.db, and supported credential dumping through Mimikatz and QuarksDumpLocalHash tools. A May 2026 activity set introduced Stowaway and TmcLoader/TmcPayload components that used network shares to exfiltrate collected data from compromised systems.

A threat actor known as ellie.191 told 404 Media they breached AI music generator Suno by compromising a single employee via the Shai-Hulud worm, then used harvested GitHub and cloud credentials to access source code, customer lists, and Stripe payment data. Leaked code revealed Suno scraped YouTube Music, Deezer, Genius, Pond5, Jamendo, Freesound, IMSLP, and roughly a million hours of podcasts, routing the scraping through Bright Data proxies. Suno confirmed the incident, dating it to November 2025 and describing it as limited, while Shai-Hulud's second wave that month added a destructive fallback wiping home directories when exfiltration fails.

Microsoft Defender Experts observed increased ACR Stealer activity between late April and mid-June 2026, with two campaigns using ClickFix social engineering to trick users into running malicious commands. The first delivers payloads via WebDAV shares, Python-based loaders, and in some cases blockchain-backed dead-drop command-and-control resolution known as EtherHiding. The second uses MSHTA, obfuscated PowerShell, and steganography that hides payloads inside JPEG images, with both chains ultimately stealing browser credentials, session tokens, and documents via DPAPI decryption.