Cybersecurity News

Google is rolling out several network-level privacy upgrades in Android 17, aimed at making encrypted web traffic harder to correlate with individual users. The headline addition is support for Encrypted Client Hello (ECH), a protocol that scrambles the Server Name Indication (SNI) during the TLS handshake. This prevents onlookers—including ISPs and Wi-Fi operators—from seeing which specific domains a user is connecting to, even when the connection itself is encrypted. Beyond ECH, the update addresses weaknesses in legacy cellular signaling. Android 17 introduces protections against IMSI catchers (often called Stingrays) and mitigates other known flaws in the mobile telephony stack that could expose a subscriber’s identity or location. On the home network side, Google is implementing changes that reduce the leakage of device-specific metadata, making it harder for third parties to fingerprint a user's local network environment. Key details from the announcement: ECH support is enabled by default in the OS-level TLS stack, though its effectiveness depends on the destination server also supporting the protocol. Cellular hardening includes tamper-resistant mechanisms for subscriber authentication, specifically targeting interception and downgrade attacks. Network privacy enhancements limit how apps and remote servers can query local network attributes, closing a vector for cross-device tracking. These changes are baked into the platform, meaning developers do not need to update their apps to benefit. However, ECH compatibility may vary across CDNs and websites that have not yet implemented the standard on their servers. Source: Unknown Are you planning any server-side adjustments to ensure your web services are ECH-compatible before Android 17 devices become widespread in your user base?

Threat research provides security teams with a clear view of how adversaries operate, but for small and medium-sized businesses, turning that knowledge into action is often the harder part. Managed Detection and Response (MDR) services bridge that gap by converting raw intelligence into continuous monitoring and faster incident response. ESET highlights that when threat intelligence is paired with human expertise, SMBs can shift from a reactive posture to a more proactive one without needing a large in-house security team. The value here lies in the combination of layers. MDR providers use threat research to fine-tune detection rules, reduce false positives, and prioritize alerts that actually matter. For SMBs lacking a dedicated 24/7 SOC, this means an external team can step in to hunt for threats, validate suspicious activity, and contain incidents before they escalate. The article also emphasizes that telemetry from the endpoint—combined with threat intelligence—allows MDR analysts to see the full attack chain rather than isolated events. Key takeaway: Threat intelligence alone is passive; MDR makes it operational. For SMBs: Outsourcing detection and response can be more cost-effective than hiring and training internal staff. Reduced alert fatigue: MDR filters out noise, letting internal teams focus only on verified threats. Faster containment: Human-led response helps stop lateral movement and data exfiltration sooner. The practical implication is that SMBs don’t need to build a security research lab to benefit from cutting-edge threat data. By subscribing to an MDR service, they gain access to the same visibility and response capabilities that larger enterprises typically deploy, but with a fraction of the overhead. ESET’s position is that this combination offers a genuine defensive edge against increasingly sophisticated attackers. Source: Unknown Is your organization currently leveraging MDR-style services, or are you still relying on internal tools alone to handle detection and response?

A newly detailed Rowhammer technique, dubbed GPUThor, is shown to bypass the error-correcting code (ECC) protections built into NVIDIA GPUs. By exploiting the physical limitations of DRAM refresh cycles, the attack can induce bit flips inside GPU memory, undermining the reliability safeguards meant to prevent such corruption. The implications are two-fold: the attack can cause a denial-of-service (DoS) condition by corrupting critical data, or it can be used for root-level privilege escalation. This undermines the long-held assumption that ECC memory is a robust defense against Rowhammer-style disturbances. Key affected configurations include GPUs where ECC is enabled by default: NVIDIA H100, A100, and A800 data center GPUs NVIDIA RTX A6000 and RTX A5000 professional GPUs Consumer GeForce RTX 3090 and RTX 4090 (where ECC is explicitly enabled) The research team demonstrated the attack without relying on any software vulnerabilities in the NVIDIA driver stack. Instead, they used a method called unprivileged GPU memory allocations, paired with massive memory traffic, to trigger the bit flips. They also confirmed that software mitigations like NVIDIA’s driver-level error reporting do not stop the attack, as the ECC correction itself is silently bypassed. For administrators, the practical risks are real but require local access or a GPU-accelerated workload. Mitigation is not straightforward: disabling ECC is not a safe alternative, and the attack appears to be resistant to current driver patches. The researchers suggest monitoring for unusual error rates in GPU logs, though this may not catch the targeted bit flips. Source: BleepingComputer Is your organization running any of the listed NVIDIA data center GPUs, and if so, what are you doing to monitor for potential Rowhammer attempts on your GPU memory?

Boston Scientific, the global medical technology firm, has confirmed that a cyberattack disrupted parts of its IT infrastructure, leading to operational interruptions across multiple regions. The company stated that the incident affected certain internal systems, though it did not specify the exact nature of the attack or the systems involved. According to the announcement, the disruption has impacted business operations worldwide, but there is no indication that patient data or clinical devices were compromised. The company has not yet released a detailed timeline or a full list of affected services. However, it appears that the attack forced Boston Scientific to take certain systems offline as a precautionary measure, which in turn caused delays in some administrative and supply chain functions. The firm is reportedly working with external cybersecurity experts to contain the incident and restore normal operations. No ransomware group has publicly claimed responsibility, and no specific CVE or advisory identifier was mentioned in the report. While the immediate impact seems limited to internal IT systems, this incident highlights the growing threat landscape for healthcare and medical device manufacturers, where operational downtime can have cascading effects. Boston Scientific has not indicated whether any third-party data was exfiltrated, and it is advising customers and partners to remain vigilant for any unusual activity related to their accounts or interactions with the company. Operational disruptions were reported globally. No evidence of compromise to patient data or medical devices has been disclosed. The company has engaged external incident response teams. No specific CVE or technical advisory has been referenced in the public statement. Source: BleepingComputer Is your organization in the healthcare or medical device supply chain, and how are you adjusting your vendor risk assessments in light of this incident?

Snowflake is officially retiring password-based authentication for legacy service accounts, pushing organizations toward passwordless access methods. While the migration itself sounds straightforward, the real difficulty—according to Token Security—lies in the discovery phase: figuring out exactly what each account is used for, who actually owns it, and whether it still requires its current level of privilege. For many teams, the first hurdle is simply inventorying these accounts. Service accounts often linger long after their original purpose has faded, and documentation is frequently sparse or outdated. Without a clear map of dependencies, removing passwords can break critical automated workflows or, worse, leave dormant accounts with excessive permissions exposed. Token Security emphasizes that the core problem isn't the password removal—it's the identity and access governance that should have been in place all along. Inventory all legacy service accounts and map them to specific applications, scripts, or integrations. Identify a responsible owner for each account—if none exists, treat it as a risk candidate. Review and reduce permissions before removing the password, ensuring least-privilege access. Migrate to passwordless alternatives such as key-based authentication or workload identity federation, where supported. Document the new authentication method and set a review cadence for ongoing account hygiene. The transition also highlights a broader industry shift: static credentials are becoming less acceptable for machine-to-machine communication. Organizations that treat this as a one-time task rather than an ongoing governance practice will likely face the same hard part again with future credential changes. Source: BleepingComputer Has your organization already completed a full inventory of its Snowflake service accounts, or are you still in the discovery phase?

Oasis Security has disclosed a weakness in NVIDIA NemoClaw that could allow a malicious webpage to take unauthenticated control of a local Ollama instance powering an AI agent. The attack goes beyond simple command execution—it can inject hidden instructions directly into the model, effectively poisoning it for future interactions. The flaw was reported to NVIDIA's Product Security Incident Response Team, with details shared ahead of public disclosure. The attack surface is notable because it does not require any prior authentication, meaning a single visit to a crafted page could compromise the integrity of the local AI model. Key technical points from the disclosure: The attack vector relies on the browser or web content interacting with the local Ollama API without proper authorization checks. Once exploited, the attacker can issue commands to the Ollama instance and modify the model's behavior persistently. The injected instructions remain embedded in the model, potentially affecting all subsequent responses generated by the AI agent. Oasis Security emphasizes that this is a supply-chain style risk for developers and users relying on local AI deployments. No specific CVE identifier was listed in the report, but the issue is understood to affect setups where NemoClaw is deployed with default or permissive network configurations. Mitigation guidance from the researchers includes restricting network access to the Ollama service, enforcing authentication for local API endpoints, and monitoring model files for unexpected changes. Source: The Hacker News Are you running any local AI models exposed to browser traffic, and if so, how are you locking down the API layer against this type of attack?

Law enforcement agencies across 22 countries have arrested 58 individuals linked to cybercrime networks coordinated by African crime groups. The operation, which involved authorities from multiple jurisdictions, led to the identification of 263 suspects in total. The suspects were connected to various cybercriminal activities, including ransomware attacks, BEC, and online scams. Investigators noted that these networks operated with a high degree of organization, coordinating attacks across borders. 58 arrests made across 22 countries. 263 suspects identified in total. Focus on networks linked to African crime groups. The crackdown highlights the growing international cooperation needed to dismantle these increasingly sophisticated operations. Law enforcement agencies have not yet released the full list of charges or the specific countries where the arrests took place. Source: Unknown Is your organization doing anything to monitor for threats that might originate from these coordinated networks?

A newly analyzed phishing-as-a-service (PhaaS) platform, dubbed AnonyMousKIT, is automating the theft of iPhone passcodes to bypass Apple’s Activation Lock on stolen devices. The service relies on voice AI agents to conduct real-time phone calls, impersonating Apple support representatives to trick victims into revealing their unlock codes. The attack chain appears to begin with traditional phishing or social engineering to collect the victim’s Apple ID credentials. Once the attacker has that information, AnonyMousKIT initiates a fraudulent call. The AI-driven voice agent convinces the target that their account has been compromised and that they must verify their identity by repeating a one-time passcode. That passcode, once harvested, is used to remove the device from the owner’s iCloud account, effectively unlocking it for resale or reuse. Key technical characteristics of the platform: Uses voice AI to mimic human interaction, reducing the need for live operators. Targets Apple Activation Lock, a security feature designed to prevent stolen iPhones from being reactivated. Relies on passcode phishing to complete the unlock process, rather than exploiting a hardware or software flaw. Operates as a service, meaning less-skilled criminals can purchase access to the tooling without building it themselves. While the article does not mention a specific CVE or Apple advisory ID associated with this campaign, the method highlights a growing trend: attackers bypassing technical security controls through social engineering rather than code-level exploits. For defenders, the key takeaway is that users should be reminded that Apple will never request a passcode or verification code over an unsolicited phone call. Source: BleepingComputer Has your organization seen an increase in AI-driven voice phishing attempts, and how are you training users to verify unsolicited phone calls?

A large distributed denial-of-service (DDoS) attack has been hammering Norway’s shared government digital infrastructure since Monday, causing outages across public sector services. The assault targets the national platform that handles authentication and secure access for citizens, including the widely used ID-portén login system. The disruption has impacted multiple agencies and municipalities that rely on this common framework for digital service delivery. While the exact scale of the traffic flood has not been publicly detailed, officials have confirmed that the attack is ongoing and that mitigation efforts are being coordinated with national security authorities. Key affected components include: Central authentication services for government portals. Secure messaging and document exchange used by public employees. Several citizen-facing applications for healthcare and tax filings. Norwegian authorities have stated that no data compromise has been detected, and services are being restored in waves as traffic filtering is applied. However, intermittent downtime is expected while the attack persists. The incident underscores how a single, shared infrastructure point can become a high-value target for disruptive campaigns. Source: Unknown Has your organization reviewed whether it relies on a shared authentication or infrastructure provider that could be similarly targeted in a widespread outage?

Attackers are shifting their focus from cracking passwords to subverting the very processes that prove who you are. The weak points are no longer just the login screen, but the identity verification workflows themselves—such as employee onboarding and account recovery. According to Specops, this is where social engineering attacks are now finding fertile ground, often resulting in fake workers being granted legitimate access to corporate systems. The core problem is that many identity verification methods rely on data that is either publicly available or easily obtained through phishing. For example, knowledge-based questions, like a mother's maiden name or a previous address, can often be answered with a simple online search. In the context of hiring, attackers may use stolen personal information to pass background checks or verification steps, creating a synthetic identity that appears legitimate on paper. This lets them slip through the cracks and become an insider with all the associated rights and access. Similarly, account recovery flows are a prime target. If an attacker can correctly answer a few "identity proofing" questions, they can trigger a password reset or take over a session. The article highlights that current systems are struggling to distinguish between a legitimate user who has lost their password and a malicious actor who has purchased or harvested enough personal data to impersonate them. To address this, the recommendation is to move beyond static, knowledge-based verification and adopt more dynamic and layered approaches. This includes using document verification, biometric checks, and device-based signals that are far harder to replicate. These methods should be applied not just at the initial login, but crucially, during the entire lifecycle of an identity—from onboarding through to account recovery. By making it harder for attackers to pass these intermediary steps, organizations can close a significant security gap that exists between the password and the network. Source: Unknown Given that knowledge-based questions and simple document checks are no longer sufficient, has your organization started using behavioral or biometric verification for account recovery, and what has the user impact been?

Researchers have shed light on a Chinese-speaking cybercrime operation tracked as UAT-10147, which is actively targeting Windows and Linux web servers across the globe. The campaign has primarily hit organizations in the education, media, technology, and gaming sectors, with the majority of victims located in Brazil, Bolivia, China, Canada, and Vietnam. The group's toolset is notable for its scale and sophistication. Attackers are leveraging artificial intelligence to automate and accelerate server compromise, allowing them to spread across a broad range of targets with minimal manual intervention. In addition to AI-driven scanning and exploitation, the operators have deployed SPECTRE, a modular backdoor designed with built-in EDR bypass capabilities to evade endpoint detection and response controls. For Linux environments, the threat actors have also deployed a custom rootkit to maintain stealth and persistent access on compromised hosts. The rootkit is likely used to hide malicious processes, files, and network connections from system administrators and security tooling. Key technical highlights from the campaign include: AI-assisted reconnaissance and exploitation to identify vulnerable web servers at scale. Deployment of the SPECTRE backdoor on Windows systems, featuring EDR evasion techniques. A dedicated Linux rootkit for stealthy persistence on compromised Unix-based servers. Focus on high-value sectors such as education and technology, with a notable geographic concentration in South America and East Asia. The findings were made public after researchers discovered an open command-and-control infrastructure linked to the group, which exposed internal operations and tooling. While full attribution remains unclear, the Chinese-speaking origins of the group suggest a highly organized operation with significant development resources. Source: The Hacker News Given the use of AI-driven automation and a dedicated Linux rootkit, how is your organization monitoring web server fleets for signs of this kind of stealthy, cross-platform intrusion?

Researchers at the University of Massachusetts Amherst have detailed a novel attack technique that can bring expired Visa contactless credit cards back to life for real-world purchases. Dubbed "Zombie Card," the method targets how point-of-sale (POS) terminals interpret data over near-field communication (NFC) rather than breaking the underlying cryptography. The core issue lies in the ability to rewrite the expiration date field that a POS terminal reads during a contactless transaction. Because the card's cryptographic authentication is not compromised, the terminal validates the transaction as legitimate, even though the physical card is past its intended validity period. This allows an attacker to perform in-store purchases using the revived card data. The attack requires physical access to the expired card or its data. The modification occurs during the NFC handshake between the card and the terminal. No cryptographic keys or security chips are broken during the process. This technique does not affect the cardholder's actual account standing, but it highlights a gap between the data a terminal accepts and the data a card issuer considers valid for payment. The researchers demonstrated that the modified card can complete transactions at standard retail POS systems that rely on contactless taps. Source: The Hacker News Does your organization's payment infrastructure validate expiration dates on the terminal side for contactless transactions, or do you rely solely on the issuer's response during the tap?

Adversa AI has disclosed a new attack technique that could let malicious web pages intercept data from xAI's Grok chatbot. Dubbed Cryptographic Context Injection, the method triggers when a user asks Grok to summarize a standard web page. In the background, the page can quietly instruct the model to exfiltrate sensitive details to an attacker-controlled server. According to the researchers, the compromised data can include the user's name, approximate location, subscription tier, and the prompts from the active conversation. The attack leverages the trust boundary between the user's browsing session and the model's processing pipeline, allowing the injected context to override expected behaviors without requiring any user interaction beyond the initial summarization request. Affected data: user identity, geolocation, plan type, and conversation prompts. Attack vector: maliciously crafted web page content processed by Grok's summarization feature. No user action needed beyond requesting a summary of the untrusted page. This disclosure highlights a broader class of prompt injection risks in consumer AI tools, where the browser becomes the attack surface. Adversa AI has not yet released a public proof-of-concept, but the implications for privacy-focused AI usage are significant. Source: The Hacker News Is your team already considering mitigations for prompt injection attacks in AI-assisted browsing tools, or is this still an unaddressed risk in your environment?

Artificial intelligence has become one of the defining technologies of this decade, reshaping industries from healthcare and finance to manufacturing and education. Organizations are increasingly leaning on AI to automate repetitive tasks, surface patterns buried in massive datasets, and accelerate decision-making. Cybersecurity is no exception—but the momentum cuts both ways. Just as defenders are exploring AI to improve their operations, attackers are using the same technology to automate and scale their campaigns. For security operations centers (SOCs), the pressure to keep up with an ever-growing volume of alerts and incidents is real. That's where platforms like Wazuh come into play. As an open-source security monitoring solution, Wazuh provides the foundation for threat detection, log analysis, and incident response. By integrating AI into those workflows, SOC teams can shift from a purely reactive posture to one that is more proactive and efficient. The practical benefits are notable. AI can assist in triaging alerts, reducing noise, and correlating events that might otherwise go unnoticed. It can also support faster investigation times, giving analysts more context and clarity when they need it most. This not only improves response times but also frees up human experts to focus on complex threats that require deep analysis. Wazuh offers a unified platform for XDR and SIEM capabilities. AI integration helps in automating repetitive SOC tasks. Pattern recognition in large datasets becomes more manageable with AI assistance. Faster decision-making is achievable when AI models support analyst workflows. The intersection of AI and security operations is still evolving, but it's clear that the ability to adapt will define how well organizations defend their environments in the years ahead. Source: The Hacker News Is your SOC team already experimenting with AI-assisted workflows, or are you still relying on purely manual triage processes?

Windows named pipes remain one of the most efficient methods for interprocess communication on the system, but they also represent a persistent attack surface when access controls are misconfigured. If a privileged service exposes a named pipe with overly permissive permissions, any local process running as an unprivileged user may be able to connect, send crafted requests, and potentially escalate privileges. This is not a theoretical concern—attackers have repeatedly abused named-pipe impersonation to hijack tokens or inject commands into higher-integrity services. The core issue often boils down to trust: developers frequently assume that only legitimate clients will reach the pipe, but on a shared or compromised host, that assumption collapses. ThreatLocker’s analysis highlights that securing named-pipe communication requires a defense-in-depth approach rather than a single fix. Endpoint verification is the first step—confirming the identity and integrity of the calling process before any data exchange occurs. This can be achieved through checks on the client’s security identifier (SID), process path, or signed executable hash. Even with that in place, command authorization is essential: the server should explicitly whitelist the set of operations it accepts, rather than blindly executing any instruction sent over the pipe. Input validation also deserves close attention, as many named-pipe exploits succeed through malformed data that triggers unintended behavior in the server-side parser. Treating all pipe input as untrusted and validating length, type, and range before use can block a significant class of attacks. Finally, narrowly scoped privileges on the service account itself limit the damage if the pipe is ultimately compromised. Running the service with the least privilege necessary—rather than as SYSTEM by default—ensures that a single flaw does not grant full control of the host. For administrators and developers, the practical takeaways are straightforward: Audit existing named-pipe ACLs to ensure only authorized users or groups can connect. Apply explicit allowlists for both client identities and permitted commands before processing. Sanitize all incoming data, enforcing strict structural and size constraints. Refuse to impersonate the connecting client unless absolutely required, and use dedicated low-privilege service accounts. Monitor for abnormal pipe activity, such as repeated failed connections or unexpected client names. Windows does not enforce secure pipe semantics by default—security is the developer’s responsibility. Hardening these channels is a necessary component of any Windows environment that relies on local services, and skipping these steps leaves a silent, local privilege escalation vector open for any attacker who has already gained a foothold. Source: BleepingComputer

Phishing attacks are increasingly powered by AI, allowing threat actors to craft highly personalized and convincing messages that slip past traditional email filters with ease. For managed service providers (MSPs), the challenge is no longer just about stopping malicious emails at the gateway—it’s about catching the ones that inevitably get through. According to guidance from Kaseya, the key is shifting focus from the inbox to the broader identity and endpoint landscape. Once a phishing email bypasses filtering, it usually triggers a chain of activity that leaves traces. Monitoring this activity is critical. To effectively detect and contain these attacks, MSPs should focus on the following areas: Identity monitoring: Watch for abnormal authentication patterns, such as impossible travel, new device enrollments, or unusual MFA fatigue attempts. Attackers often use stolen credentials to access cloud apps. Email activity analysis: Look for post-delivery anomalies like mailbox rule creation, automatic forwarding to external addresses, or unexpected replies to internal threads. These are classic signs of a compromised account. Endpoint behavior: Track for suspicious processes spawned from email clients, unusual PowerShell usage, or the staging of files in temporary directories. These often indicate the execution of a payload. Cross-referencing signals: The real power lies in correlating a suspicious email with subsequent login attempts or file access patterns. A single alert might be false positive, but a chain of events is a strong indicator of compromise. The threat isn't just the initial click—it's what happens afterward. Since filters will never be perfect, continuous visibility into user behavior and endpoint telemetry provides the last line of defense. By adopting this layered monitoring approach, MSPs can contain threats that would otherwise linger undetected in the environment. Source: BleepingComputer

Title: OpenAI Halts Frontier Model Training to Strengthen AI Safety Protocols Summary: OpenAI has temporarily paused reinforcement learning training for its frontier AI models to implement enhanced safety measures and expand monitoring scope, reducing risks of unsafe internal AI behavior. Body: OpenAI announced this week that it has paused reinforcement learning (RL) training for its most advanced AI models for a two-week period. The decision was made to allow the company to implement additional defensive safeguards and widen its monitoring capabilities, following a prior incident involving the Hugging Face platform. The company stated that as models grow more capable, the risks associated with internal development and testing also increase. This pause was initiated to prevent similar lapses and to reinforce oversight during active training phases. During this period, OpenAI is focusing on improving real-time detection of unsafe model behaviors and tightening control mechanisms within its training environment. The goal is to ensure that frontier models remain within predefined safety boundaries before resuming full-scale RL training. The original article, published by The Hacker News, provides further context on the incident and the company's broader safety strategy. You can read the full report here: https://thehackernews.com/2026/08/openai-pauses-frontier-rl-training-as.html

Title: Spectre Attack Against Cloudflare Workers Demonstrates Remote JWT Theft at 12 Bits Per Second Summary: Researchers have published a proof-of-concept attack demonstrating a remote Spectre side-channel against Cloudflare Workers, successfully extracting a JSON Web Token from a co-located Worker in a production environment at a rate of up to 12 bits per second—a significant improvement over previous research from 2021. Body: A newly disclosed proof-of-concept demonstrates a successful remote Spectre attack targeting Cloudflare's Workers platform. The research team was able to extract a JSON Web Token (JWT) from a victim Worker running in the same environment as an attacker-controlled Worker, all within the production infrastructure. This new attack achieves a data exfiltration rate of up to 12 bits per second. This represents a 360-fold increase in transmission speed compared to an earlier Spectre-based attack against the same platform, which was demonstrated in 2021. The ability to remotely leak sensitive credentials like JWTs from co-located workloads poses a significant concern for multi-tenant serverless environments. The end-to-end experiment was conducted using two Worker instances: one serving as the attacker and the other as the victim. The attack was designed and executed solely by the researchers to validate the vulnerability and measure its practical impact. Details on the specific exploitation techniques and the attack chain have been published in the official disclosure. This research highlights the ongoing challenges of securing shared cloud infrastructure against speculative execution side-channel attacks. While microarchitectural mitigations exist, verifying their effectiveness in dynamic edge computing platforms remains a complex and ongoing task. For further details, please see the original report by The Hacker News. Source: The Hacker News URL: https://thehackernews.com/2026/08/cloudflare-workers-spectre-attack-leaks.html

Title: The Next Phase of Phishing: AI Agents Are Now Attacking and Defending Inboxes Summary: Traditional email defenses that rely on scanning for malicious links or attachments are no longer sufficient. The threat landscape has shifted from malicious payloads to malicious intent, and with the rise of AI agents on both sides, the battlefield has moved to an agent-versus-agent paradigm. Body: The core methodology of most email security tools has remained largely static for the better part of a decade. The standard approach involves scanning incoming messages for known signatures, bad links, or dangerous attachments. This strategy was effective when the threat resided solely in the payload. However, this model is breaking down as cybercriminals shift the focus from what is in the message to what the message intends to achieve. The landscape is now evolving further, as the threat actor on the other end of the conversation is no longer a human being, but an autonomous AI agent. This evolution marks a transition through distinct phases of phishing. The earliest iterations, often referred to as Phishing 1.0, relied on crude social engineering and easily detectable malicious content. While effective in their time, these attacks were relatively straightforward for security filters to identify and block. The current shift signifies a move away from static detection methods toward understanding the semantic context of a conversation. As AI agents become capable of conducting entire email exchanges, simulating human behavior and reasoning, they complicate the ability of traditional security tools to distinguish between legitimate business requests and malicious manipulation. Defenders must now fight fire with fire, deploying their own AI agents to analyze intent and behavior rather than just parsing code, signaling a new era where the success of an attack or defense depends on the sophistication of the algorithms on either side of the inbox. Source: The Hacker News Original Article: https://thehackernews.com/2026/08/phishing-30-fight-moves-to-agent-versus.html

Title: Operation CameraSwarm: Over 14,500 Dahua Devices Compromised via Credential Attacks and Auth Bypasses Summary: Threat actors compromised more than 14,530 Dahua video surveillance devices in a campaign spanning from June 17 to July 22, 2026. The operation, tracked as Operation CameraSwarm, leveraged credential stuffing, two distinct authentication-bypass vulnerabilities, and a peer-to-peer (P2P) relay technique to gain unauthorized access. The findings were reconstructed from a 407 MB exposed working directory containing 2,616 files. Body: Researchers at Hunt.io have published a detailed analysis of a significant attack wave targeting Dahua devices. Between June 17 and July 22, 2026, at least 14,530 units were successfully compromised. The attackers employed a multi-pronged strategy combining brute-force credential attacks with exploitation of two separate authentication-bypass flaws, followed by the use of a P2P relay technique to maintain access and obfuscate command-and-control traffic. The investigation was made possible by the discovery of a 407 MB working directory left exposed by the threat actors. This directory, containing 2,616 files, allowed researchers to reconstruct the entire kill chain and attribute the activity to a campaign now codenamed Operation CameraSwarm. While the specific CVE identifiers for the authentication-bypass flaws were not disclosed in the initial reporting, the combination of weak credential reuse and unpatched vulnerabilities is cited as the primary vector for the initial compromise. The use of the P2P relay method is particularly noteworthy, as it leverages the devices' legitimate remote-access features to blend malicious traffic with standard operational data, making network-based detection more difficult. This incident underscores the critical importance of changing default credentials, enforcing strong password policies, and ensuring firmware is updated to address known authentication bypasses. Organizations using Dahua equipment should immediately audit their device exposure and review P2P settings. Source: The Hacker News Original Article URL: https://thehackernews.com/2026/08/hackers-compromised-14500-dahua-devices.html