DeepSeek AI weaponized for autonomous server attacks

Monday Free Edition - August 3, 2026

Threat of the Week: DeepSeek AI Deployed for Autonomous Cyberattacks

A Chinese-speaking threat actor has been observed using the DeepSeek AI model paired with the open-source Hermes Agent to conduct autonomous cyberattacks against exposed servers — with minimal human involvement required. According to Bleeping Computer, the attacker leverages DeepSeek's reasoning capabilities to identify, target, and exploit vulnerable systems at a speed and scale that would be difficult to achieve manually.

This represents a meaningful shift in the threat landscape: rather than AI being used merely to assist in drafting phishing emails or generating code snippets, it is now being used as an operational engine that drives the attack chain end-to-end. The use of an open-source agent framework like Hermes lowers the barrier to entry significantly, meaning this approach is likely to be replicated by other actors once techniques are refined and documented in underground communities.

Why it matters: Autonomous AI-driven attacks compress the time between initial reconnaissance and exploitation, giving defenders less opportunity to detect and respond. Organizations with internet-exposed services — unpatched web applications, legacy APIs, or misconfigured cloud endpoints — are the most immediately at risk. Prioritize attack surface reduction and ensure exposed services are inventoried and patched.

Deep Dive: ESET's Threat Report Flags Malicious AI Skills and Evolving Malware

The DeepSeek story does not exist in isolation. ESET's latest threat report, covered by Bleeping Computer, documents a broader and accelerating trend: attackers are systematically adapting established techniques to AI platforms, emerging technologies, and shifting user behavior.

Key findings from the ESET report include:

Malicious AI skills: Threat actors are building and distributing skills or plugins for AI platforms designed to harvest data or facilitate further compromise.
AI-assisted malware: Malware is increasingly being developed or refined with AI assistance, enabling faster iteration and more adaptable payloads.
ClickFix attacks: Social-engineering techniques that manipulate users into executing malicious commands are on the rise, exploiting trust in familiar interfaces.
Record quishing activity: QR code phishing (quishing) reached record levels during the reporting period, as users remain less suspicious of QR codes than traditional links.
Ransomware security-disabling tools: ESET observed ransomware tooling specifically designed to disable or circumvent endpoint security software before payload deployment.

Taken together with the DeepSeek autonomous attack story, the ESET report paints a picture of an ecosystem in which AI is no longer a novelty in the attacker's toolkit — it is becoming a standard component. Defenders should treat AI-augmented threats as a baseline assumption rather than an edge case when modeling adversary capabilities.

Hack of the Week: Adform Ad Script Compromised in Cryptocurrency Supply-Chain Attack

Online advertising firm Adform suffered a supply-chain attack in which malicious actors compromised one of its scripts to deliver cryptocurrency-stealing code to websites using its ad platform. As reported by Bleeping Computer, the attack used a technique known as clipboard hijacking: when visitors to affected sites copied a cryptocurrency wallet address, the malicious script silently replaced it on the clipboard with an address controlled by the attacker.

This attack is notable for several reasons. First, it exploited the trusted relationship between a legitimate advertising platform and the publishers and users who rely on it — a classic supply-chain vector. Second, clipboard hijacking is an effective technique against cryptocurrency users because the long, opaque format of wallet addresses means victims rarely notice the substitution before completing a transaction. By the time a victim realizes funds have been misdirected, the transaction is irreversible.

Takeaways for organizations and individuals:

Businesses embedding third-party scripts should implement Subresource Integrity (SRI) checks where possible, and audit third-party JavaScript dependencies regularly.
Cryptocurrency users should always manually verify wallet addresses character-by-character before confirming any transaction, rather than relying solely on pasted values.
Security teams should consider Content Security Policies (CSP) that restrict which external scripts can execute on company-owned web properties.

Tool Spotlight: CISA Warns on PLC Attacks Targeting Water Utilities

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has issued a warning about a significant increase in cyberattacks targeting internet-exposed programmable logic controllers (PLCs) in the water and wastewater systems sector, according to Bleeping Computer. These attacks are causing operational disruptions at water utilities across the United States.

PLCs are the industrial control hardware that manage physical processes — in water utilities, they govern pumps, valves, chemical dosing systems, and treatment operations. When attackers gain access to internet-exposed PLCs, the consequences extend beyond data loss into the realm of physical operational disruption, and in worst-case scenarios, public safety risk.

CISA's warning highlights a persistent and well-documented problem: critical infrastructure operators frequently expose operational technology (OT) devices to the public internet either inadvertently or for remote management convenience, without adequate authentication controls or network segmentation. Key mitigations emphasized by CISA and consistent with established OT security guidance include:

Remove PLCs and other OT devices from direct internet exposure; place them behind firewalls and VPNs.
Implement multi-factor authentication for any remote access to industrial control systems.
Segment OT networks from IT networks to limit lateral movement in the event of a breach.
Conduct regular asset inventories to identify and remediate unintentionally exposed devices.

This advisory is a reminder that the attack surface for critical infrastructure is not abstract — it is measurable, often discoverable via public scanning tools, and actively being targeted. Water sector operators of all sizes should treat this warning as an immediate prompt to audit their internet-facing OT exposure.

Breach Board: This Week's Notable Incidents and Disclosures

A summary of additional developments covered in this issue and worth tracking:

Adform supply-chain compromise (Consumer/Finance): Cryptocurrency clipboard-hijacking script delivered via compromised ad platform script. Affects any website visitor who copied a wallet address while on an impacted publisher site. Source: Bleeping Computer.
U.S. water utilities PLC disruptions (Critical Infrastructure/Government): CISA confirms active attacks disrupting water and wastewater operations via internet-exposed PLCs. Source: Bleeping Computer.
DeepSeek-powered autonomous server attacks (Enterprise): Chinese-speaking threat actor using AI and open-source agent tooling to attack exposed servers with limited manual intervention. Source: Bleeping Computer.
ESET Threat Report — record quishing and ransomware tool evolution (Enterprise/Consumer): Broad threat trend data flagging AI skills abuse, ClickFix, clipboard attacks, and security-disabling ransomware tooling. Source: Bleeping Computer.

On the radar (industry/vendor): OpenAI announced price reductions for two of its GPT-5.6 models — cutting Luna's API price by 80% and Terra's by 20% — as it continues to drive cost efficiency in its model offerings. While not a threat story, falling AI inference costs have direct implications for attacker economics: the same cost reductions that benefit legitimate developers also make AI-augmented attack tooling cheaper to operate at scale. Source: Bleeping Computer.