Week in review: Cisco fixes IMC bug, Patch Tuesday forecast, Black Hat USA 2026

Here’s an overview of some of last week’s most interesting news, articles, interviews and videos:

Mapping the malware blast radius a single alert won’t show you
In this interview with Help Net Security, Mike Wiacek, founder and CTO of Stairwell, explains Backstory, an AI agent that takes a single alert and works outward to map how far a malware campaign spread. He walks through the research behind the claim that each published sample hides an average of 2.4 undocumented variants, describes what counts as a related variant, and explains why Stairwell keeps every executable that runs on customer endpoints.

Elastic Defend now covers 800+ vulnerable drivers, with automated troubleshooting and ARM support
Attackers reaching for kernel access on a Windows machine bring a driver Microsoft already trusts. It is signed, it loads, and it carries a known flaw. That flaw gives them enough room to tamper with memory or disable the security software watching the host.

Analysts got 19 minutes back every hour in Stellar Cyber’s agentic auto triage trials
An analyst opening a queue on Monday morning will spend most of it on tickets that amount to nothing. Stellar Cyber’s Agentic Auto Triage closed 8,047 of those tickets on its own during customer trials, filing them as confident false positives.

OWASP’s subtractive security project measures the attack paths you erased
An attacker who talks a user into opening an attachment gets whatever that machine still permits: a service account with rights across the domain, an outbound route to anywhere, a scripting engine sitting there for the taking. Christopher Frenz wants those capabilities deleted before anyone writes a detection rule for them.

Digital executive protection is a strategic imperative for CEOs
In this interview with Help Net Security, Brian Hill, Field CISO, Client Advisory for BlackCloak, explains how attackers reach companies through the personal lives of executives. He describes a case where a draft report sat in an executive’s personal email with no multifactor authentication, and traders acted on it before the news went public.

What stops attackers wrecking industrial plants is knowing how
Engineers at an Israeli food producer spent most of a week rebuilding a refrigeration system after an intruder switched the gas cooler and receiver valves to manual and pinned them open. Liquid CO2 flooded the compressors and destroyed them. The replacement units did not match the originals, so the whole system had to be reworked and recharged with gas. That incident is one of roughly forty in Kaspersky ICS CERT’s quarterly roundup of attacks on industrial organizations covering the second quarter of 2026.

15 TP-Link Omada vulnerabilities let attackers hijack routers and intercept camera traffic
TP-Link prints Omada router serial numbers on the device and packaging. Because the serials are sequential, guessing one and querying the Omada cloud reveals the device’s MAC address and model. For example, serials starting with 22460J500 map to ER605 routers, while 224608100 identifies ER7206 models. Forescout’s Vedere Labs used this behavior as part of an attack chain, one of several built from 15 vulnerabilities affecting TP-Link’s Omada networking platform.

Code review used to be the only way to catch these bugs
An automated system called NOVA read the source code of 3,915 open-source projects over two months and came back with 14,090 vulnerabilities, each one confirmed through the system’s validation pipeline. Vulnerability researchers at Palo Alto Networks’ Unit 42 built the system and checked its output against the public record afterward.

Pre-auth RCE in enterprise Java hits Bonita and OFBiz servers
A single unauthenticated web request let attackers reach an internal Bonita API and execute code on the server. The flaw affected Bonita BPM, widely used to automate workflows in banking, insurance, and government. Researchers at Novee discovered the pre-auth remote code execution vulnerability and presented their findings at Black Hat USA 2026.

Cloudflare OS goes open source with a record of everything its agents read
Cloudflare open sourced Cloudflare OS, the AI agent platform its employees had been using internally since May. The platform tracked every resource an agent accessed and carried that context into the agent’s outputs. When someone opened the output, Cloudflare OS verified they had permission to access the underlying data before revealing it. For example, a dashboard built from a sensitive database table remained hidden from users who lacked access to that table.

August 2026 Patch Tuesday forecast: How do we deal with the patch apocalypse?
July 2026 Patch Tuesday was record-setting in so many ways. The sheer volume of security patches for almost every product in the Microsoft portfolio was the highest ever and, of course, well over 600 CVEs were identified in the Security Updates Guide.

ShieldFont fights AI scraping by handing crawlers the wrong words
Isaque Seneda and Gabriel Abrucio built a web font that draws one set of words on screen and leaves a different set in the page’s source code. A person reading in a browser sees the writing as written. A scraper pulling the HTML gets different words in the same grammar, at the same URL, off the same bytes.

What the first year of EU AI Act transparency enforcement could look like
In this Help Net Security interview, Edwin Weijdema, Field CTO at Veeam, answers questions on Article 50 of the EU AI Act and what the first year of enforcement might bring.

Browser security is where software, data, and AI meet
In this interview with Help Net Security, Rui Ribeiro, CEO of Jscrambler, explains why the browser has become a security problem organizations do not control. Companies do not own the device, the extensions, or the network path, yet that is where application logic, third-party code, customer data, and AI meet during every customer interaction.

Three in four AI-generated vulnerability patches leave something broken
Ask a frontier model to patch a real vulnerability and it often produces code that looks convincing. It resembles a maintainer’s fix and may even pass tests, but researchers at 1Password found that only about one in four patches for six newly disclosed CVEs actually fixed the vulnerability. The rest often addressed only part of the issue or introduced new flaws.

200 accounts compromised in Swiss government’s Microsoft SharePoint breach
Hackers exploited vulnerabilities in Microsoft SharePoint servers belonging to Switzerland’s Federal Office of Information Technology, Systems and Telecommunication (BIT), compromising the login credentials of around 200 accounts.

OpenAI drops ChatGPT text chat limits for free users, adds new safeguards for teens
OpenAI has updated GPT-5.6 Sol, the model behind ChatGPT for Plus and Pro subscribers, and pushed a new model, GPT-5.6 Luna, out to everyone using the free tier.

KindaRails2Shell threatens Ruby on Rails apps (CVE-2026-66066)
A critical security vulnerability (CVE-2026-66066) in Ruby on Rails (aka Rails), one of the most widely used frameworks for building websites and web apps, may allow attackers to read sensitive files off a server and, in some cases, take full control of it.

Attackers exploit N-able N-central flaw to reach managed endpoints (CVE-2026-18577)
Attackers are exploiting an authentication bypass vulnerability (CVE-2026-18577) in N-able N-central, a remote monitoring and management (RMM) solution widely used by managed service providers, to gain access to managed endpoints.

Bank of America impersonators weaponize ScreenConnect, then make it hard to remove
A phishing campaign impersonating Bank of America (BoA) is underway, trying to trick Windows users into installing ScreenConnect remote access software and then making it difficult to uninstall it.

AI agent deception moves from theory to reality in UK cyber tests
“During a routine cyber evaluation, AI agents took sustained, unsanctioned action directed at real people and organisations,” UK’s AI Security Institute (AISI) disclosed on Tuesday. The agents’ actions included an attempted supply-chain attack that saw them create malicious pull requests and try to socially engineer an open-source maintainer into approving the malicious code (they refused).

OWASP 2026 LLM Top 10: “The model will be fooled”
The OWASP GenAI Security Project has released the 2026 edition of its Top 10 for LLM Applications and, for the first time, the list was influenced by real-world incidents. The two top entries – Prompt Injection and Sensitive Information Disclosure – remained constant, but the order shifted more than in past years below them.

Critical Cisco IMC bug gives attackers root, PoC is out (CVE-2026-20200)
Cisco has fixed a critical vulnerability (CVE-2026-20200) in its Integrated Management Controller (IMC), which allows an attacker to run commands as root through the controller’s web interface. The fix was part of Cisco’s August 5 advisory batch, and unlike the bugs squashed by the hardening releases for IOS XE and SD-WAN, this one has a public proof-of-concept exploit.

AI cut phishing from hours to seconds, which is where DMARC and BIMI come in
In this Help Net Security video, Mike Boyle, VP of Business Units at GMO GlobalSign, and Rahul Powar, CEO and founder of Red Sift, unpack the evolution of email security and why it matters for business trust.

Photos: Black Hat USA 2026 Arsenal
Help Net Security was at Mandalay Bay, where Arsenal ran alongside the Briefings. If you’ve never been, it’s the part of Black Hat that felt least like a conference and most like a workshop: a room full of stations where the people who built the tools stood behind laptops and demonstrated them. Everything on display was open source, and nearly all of it was available to download immediately.

Chinese hacker used DeepSeek to launch autonomous cyberattacks on vulnerable servers
A Chinese threat actor operating under the aliases “knaithe” and “KnYuan” used multiple LLMs to automate cyberattacks against internet-facing systems with limited human intervention. Researchers at Palo Alto Networks’ Unit 42 uncovered the operation after the threat actor’s AI agent misconfigured a file server, inadvertently exposing the entire infrastructure.

Fake IRS letters direct crypto holders to bogus compliance portal
Scammers are sending physical letters to cryptocurrency holders that copy the look of official IRS notices. The letters tell recipients they must enroll in something called a Digital Asset Compliance Portal before a deadline, or risk penalties.

EU begins enforcing AI Act, putting AI models under the microscope
Europe’s fight to regulate AI models moved from paper to practice on 2 August 2026, when the European Commission’s AI Office and national authorities began enforcing the AI Act. On the same date, new transparency rules took effect, requiring certain AI systems to tell users when they’re interacting with AI and when content has been generated or altered by it.

Russian hackers abuse hotel Wi-Fi networks to steal Microsoft 365 credentials and deploy malware
Midnight Blizzard, the Russian threat actor tied to the country’s foreign intelligence service, has spent months targeting users of public Wi-Fi networks at places like hotels and conference centers, according to new findings from Microsoft Threat Intelligence.

AI developers targeted via trojanized GitHub repositories
Cybercriminals are cloning popular GitHub repositories for AI tools and developer resources to distribute an infostealer, according to Netskope Threat Labs. Netskope came across the campaign while tracking a Windows-based MaaS infostealer, first reported in April 2026, that was spread through the ClickFix social engineering trick.

Discounted Claude access bought on the gray market may expose every prompt you send
More than half a dozen services advertised on underground forums and messaging platforms, offering discounted or “unlimited” token access to frontier AI models, were discovered by Okta. The trend is likely driven by Chinese users seeking access to AI models that are unavailable because of regulatory and provider restrictions.

Snowflake hacker pleads guilty, faces up to 32 years in prison
A Canadian man is facing decades in prison for hacking customer accounts at cloud storage provider Snowflake and stealing data from more than 165 organizations.

Buying TikTok followers can expose users to scams and account theft
Buying TikTok followers, likes, or views could do more than inflate engagement metrics. According to Malwarebytes, many services selling social media growth operate through deceptive practices that can expose customers to scams, stolen accounts, and financial loss. The market for artificial social media engagement also creates security risks for both buyers and other platform users.

Product showcase: Guardio Mobile Security turns breach alerts into a recovery plan
Guardio Mobile Security brings several protection features to iPhone and Android, allowing users to monitor exposed personal information, identify phishing attempts, and receive alerts about emerging threats from a single application. It is available on smartphones and tablets, while browser extensions support Chrome and Microsoft Edge on laptops and desktop computers.

SkillSpector: NVIDIA’s open-source security scanner for AI agent skills
SkillSpector is an open-source scanner from NVIDIA that reads an agent skill and tells you whether to install it. Point it at a directory, a zip file, a single SKILL.md, or a Git URL, and it returns a list of findings, a risk score, and recommendations.

OpenAI reveals how criminals used ChatGPT to run scams
OpenAI banned a coordinated network of ChatGPT accounts that likely originated in Cambodia’s Preah Sihanouk province, a region reports have linked to online scam compounds and human trafficking operations. The network used the company’s models to create and manage fake online personas, generate and translate messages sent to scam targets, produce promotional content for fraudulent schemes, and support daily operations.

Qodana 2026.2 adds post-quantum crypto checks for JVM code
Qodana 2026.2 shipped with new security inspections, published benchmark results, post-quantum cryptography checks, and coverage reporting that no longer has to be pointed at the reports.

CISA lays out new guidance for using open-source software
The US Cybersecurity and Infrastructure Security Agency (CISA) has published the Open Source Software: Security Principles and Practices guide, which provides federal agencies with recommendations for managing the security of open source software, contributing to OSS projects, and evaluating open source AI systems.

Microsoft shortens NuGet API key lifetime to improve supply chain security
Microsoft is reducing the lifetime of new NuGet.org API keys from 365 days to 30 days starting August 17, 2026, to improve the security of NuGet, its package repository for .NET developers. API keys created before August 17 will remain valid until November 1, after which developers will need to generate new keys or switch to NuGet Trusted Publishing.

Uptime Kuma 2.5.0 waits two weeks before trusting a new npm package
Uptime Kuma checks whether a website, a Docker container, a DNS record, or a Steam game server is still answering, and pushes a message to Telegram, Slack, or email when one stops. The change with the widest reach in version 2.5.0 is that the project now sets a 14-day cooldown on npm updates to minimize supply-chain attacks.

Your enterprise AI footprint is about three times bigger than your model list
Organizations are building AI systems that combine models, agents and external tools instead of relying on standalone AI, according to Snyk’s latest State of Agentic AI Adoption report. The study analyzed 3,044 enterprise environments and 1.39 million code repositories to examine how enterprises are deploying AI.

Future AGI: Open-source platform for shipping self-improving AI agents
Future AGI is an open-source platform for tracing, evaluating, simulating, and guardrailing LLM agents, licensed Apache 2.0 and self-hostable. Self-hosted instances register with Future AGI on first boot and send an instance ID, a version string, a deployment type, and the email addresses and domains of active admin users.

Cloudflare gives AI agents wallets with built-in spending controls
Cloudflare’s Wallets will give AI agents running on its platform a human-readable wallet handle for paying APIs and online content within limits set by their creator. Handle reservations have opened, while the service will become available in the coming months. It will include two types of digital wallets: Account Wallets and Virtual Wallets.

INTERPOL flags AI as the new engine of African cybercrime
Africa’s growing digital economy is exposing governments, businesses and internet users to a rising wave of cybercrime. The continent recorded more than 1.1 billion mobile subscriptions and over $1.1 trillion in digital transactions in 2025, while more than 570 million people relied on the internet for banking, government services, healthcare and education.

Top product launches at Black Hat USA 2026
Black Hat USA 2026 took place in Las Vegas, where vendors used the event to unveil technologies they hoped would shape the next year of cybersecurity. Here were the announcements that drew the most attention, and why they mattered for teams evaluating new security investments.

Non-human identities are 91% of everything active in production
A backup job fires at two in the morning. A scanner walks the same AWS account an hour later, a deployment pipeline assumes a role at four, and a logging agent runs straight through the night. Each of those actions carries a credential issued to a machine.

Suppliers, logins, and AI tools are all becoming attack paths
Cybercriminals and state-backed hacking groups are abusing trusted identities, cloud services, AI tools, and software supply chains to gain access while avoiding detection, according to CrowdStrike’s 2026 Threat Hunting Report.

Microsoft extends zero trust deeper into enterprise AI
Microsoft expanded its Zero Trust for AI strategy with updates to the Zero Trust Assessment tool and the Zero Trust Workshop. The additions help organizations assess security posture, prioritize remediation, and apply zero trust principles to AI agents and AI-assisted software development.

Product showcase: Material unifies Google Workspace email, file & OAuth defense
Here’s an uncomfortable truth: the attack that gets you probably won’t look like an attack. It’ll look like a normal login, file share, or permission request. Phishing, data loss, and OAuth abuse aren’t isolated problems, they’re different ways attackers exploit the same cloud workspace. Material Security secures Gmail, Drive, identities, and third-party apps as one surface, not four separate ones.

Cybersecurity jobs available right now: August 4, 2026
We’ve scoured the market to bring you a selection of roles that span various skill levels within the cybersecurity field. Check out this weekly selection of cybersecurity jobs available right now.