ACTIVE THREATS: Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Claude Distillation Attacks • Artifactory flaws chained in attacks deploying backdoor malware • Passkey-themed phishing attacks lead to Microsoft 365 data theft • Florida confirms DMV database breached via stolen police account • How Threat Actors Are Turning Trusted AI Platforms Into an Attack Surface ACTIVE THREATS: Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Claude Distillation Attacks • Artifactory flaws chained in attacks deploying backdoor malware • Passkey-themed phishing attacks lead to Microsoft 365 data theft • Florida confirms DMV database breached via stolen police account • How Threat Actors Are Turning Trusted AI Platforms Into an Attack Surface
The full HackWire archive — 3,862 stories, newest first.
18 malicious npm packages posed as Alibaba Cloud tools, using typosquatting to deliver a cross-platform remote access trojan. The targeted attack exploited developer trust through postinstall scripts, risking full machine compromise.
Liechtenstein's beneficial ownership register—a database exposing who controls shell companies—was breached, compromising hidden financial relationships and ownership structures. Unlike typical data theft, this information doesn't depreciate; it's permanent intelligence with indefinite strategic val
Visa acquired behavioral intelligence firm BioCatch for $2.4B to detect fraud by analyzing how users type, hold phones, and interact with devices—spotting patterns that reveal coaching or deception. This targets authorized push payment scams where victims' legitimate credentials are valid, but their
Black Hat 2026 has become a vendor announcement spectacle rather than a research conference, with 40+ companies launching products simultaneously in clustered categories (AI, IAM, cloud security). The resulting marketing noise drowns out genuine innovations, making it difficult for defenders to iden
Fake Roblox cheat tools on YouTube/Discord deliver malware to young players, stealing passwords and wallets. Kids seeking exploits disable security protections, making them easy targets for attackers deploying infostealers and remote access trojans.
Russian hackers use DOUBLECUP: a loader that tricks users via ClickFix social engineering, then hides malware in PNG images in browser cache—appearing as normal traffic to security tools.
A Chinese operator weaponized DeepSeek in the first documented AI cyberattack targeting Jesta Security. The agent scanned 1,200+ hosts for proxyjacking before security researchers seized control.
Attackers exploited SonicWall zero-days for six weeks to steal MFA seed credentials—enabling unlimited authentication cloning. This pre-attack reconnaissance preceded INC Ransomware deployment against organizations across continents.
Malware can bypass Chrome passkey protection by extracting the device key and disabling the User Verified flag. The vulnerability's impact depends entirely on whether websites check this flag—some sites like GitHub enforce it, while others like eBay initially didn't.
N-able's incomplete patch leaves CVE-2026-18577 unpatched, enabling administrative account takeover of N-central. Since MSPs use this platform to manage hundreds of downstream clients, a single compromise affects entire client rosters—repeating the Kaseya VSA supply-chain attack pattern.
Four breaches—poisoned AI models, Bitcoin theft, water/DNS attacks—exposed eroded trust. The AI supply chain lacks security controls, enabling code execution via compromised model downloads.
BTMOB, an Android remote-access trojan launched in 2025 as a premium paid service, has fractured into an uncontrolled secondary market of resellers, impersonators, and independent operators after server outages crippled the original provider. The malware kit's loss of control accelerated rather than
ExfilSquad breached England's Police National Legal Database, exposing 100,000+ officers' names and emails. Unlike typical data leaks, this endangers armed professionals with real enemies. The group demands ransom.
CISOs burn out in ~2 years from overload. Organizations stacked legal liability, regulatory demands, and board accountability into one position. Perks won't fix it—the job needs complete redesign.
INC Ransomware is actively exploiting SonicWall SMA1000 appliances to gain root access at enterprise network edges, enabling direct internal penetration before endpoint protections can engage. This perimeter compromise is particularly dangerous because attackers bypass EDR entirely and operate insid
N-able's N-central RMM vulnerability was patched, then bypassed by attackers. Since MSPs use it to manage hundreds of downstream clients, a single compromise can spread across entire supply chains.
Horizon3's $250M funding validates a market shift from traditional point-in-time pentesting to continuous autonomous testing. Their NodeZero platform repeatedly attacks your own network, finding vulnerabilities and misconfigurations before real attackers do.
River Bank disclosed a June ransomware attack that exfiltrated customer data, with attackers claiming deletion—but such promises lack cryptographic proof and are meaningless. This reflects the modern double-extortion model where criminals hold both encryption keys and stolen data to threaten publica
Brinks Home exposed 4.9M customer records when ShinyHunters breached their CRM, stealing 41GB of data. Though alarm systems weren't compromised, criminals now have customers' addresses and know they own valuables—the real threat isn't system takeover, but targeted physical crime.
U.K. police database breached, exposing 108K+ officers' contact info and public inquiries. ExfilSquad claimed it. Configuration failure matches a pattern hitting dozens of organizations.
The iOS exploit kit DarkSword, originally state-sponsored, is now deployed by Chinese threat actors using GHOSTBLADE malware to steal iPhone credentials. The source code leak democratized access to the tool, transforming a nation-state weapon into a commodity threat available to opportunistic actors
Organizations misunderstand two AI types in security: collaborative tools (like Claude) assist analysts on specific tasks, while autonomous systems triage alerts. Confusing them is expensive.
Thermo Fisher patched a vulnerability in its DNA sequencing software that allowed attackers to modify genomic files while evading integrity checks, potentially compromising forensic evidence, clinical diagnoses, and drug development trials without detection.
Iran-linked actors compromised US water utilities across 7+ states via vulnerable cellular routers on industrial control systems. This known attack vector—previously used against Israeli facilities in 2020—remains unpatched, leaving smaller utilities with limited IT staff highly exposed.