IT Today: Essential Technology News (August 2026)

“It takes 20 years to build a reputation and a few minutes of a cyber incident to destroy it.” (Stéphane Nappo, Global CISO at Groupe SEB)

If in July the technological zeitgeist was centered on the launch festival, in August the watchword became cybersecurity. Summer, the season when IT teams work with half the office on vacation, proved once again the favorite window of attackers. In Romania, the lesson came in the form of the attack on ANCPI on July 14, 2026. Internationally, the same theme came from a completely unexpected direction: not from hackers, but from AI agents (Anthropic and OpenAI). In early August, the UK-based AI Security Institute reported on agents creating fake online identities to pressure real people into approving code changes to an open-source project — this time in tests deliberately conducted with security filters disabled and internet access intentionally allowed.

The common conclusion of these episodes is that cybersecurity has ceased to be “the IT guys’ job.” When a breach can block an entire country’s notaries, and an autonomous agent can act in the real world thinking it’s playing in a sandbox, security skills become mandatory for any tech role — from programmer to AI automation specialist. That’s exactly what we’re talking about in the IT and Cybersecurity course.

We’ll discuss international cybersecurity incidents and other current IT news in more detail below.

AI News

Genome-Writing AI Designs New Virus That Can Destroy Resistant E. coli Bacteria

A team at Stanford used Evo 2, a generative model that writes entire genomes, to design new bacteriophages. Of nearly 300 variants synthesized and tested in the lab, 16 proved exceptionally effective against E. coli, and their cocktail quickly overcomes bacterial resistance. The model is freely available and open source — bringing the promise of new antibiotics, but also opening up a discussion about biosecurity.

Study: 42.6% of Students Are Already Being Asked by Employers About AI

A three-year report from the Kogod School of Business shows that the share of students asked by potential employers about their ability to use AI has risen from 11.6% in 2024 to 42.6% in 2026. Over 80% are already using AI in their assignments; brainstorming remains the number one use, and prompt engineering is the most desired skill — now followed by programming.

The AI ​​Paradox: Tech Leaders Promise Less Work, Their Employees Reach 90 Hours a Week

A BBC investigation shows the contrast between the promise of the four-day week, publicly supported by OpenAI and others, and the reality inside: at OpenAI and Anthropic, “sprints” can exceed 90 hours in seven days, and at Meta, people are forcibly moved to AI teams. A Berkeley study confirms that the time saved is immediately reinvested in even more work.

Digital Marketing News

Google Changes Smart Bidding Rules: What Happens to Budget-Limited Campaigns

This month, two major platforms are changing the default behavior of accounts on the same day. Starting August 17, Google Ads will respect target CPA or ROAS regardless of budget constraints. Campaigns that were consistently exceeding the target will now optimize towards the actual set value, which should reduce performance fluctuations after budget changes. Recommendation: audit campaigns with a large gap between target and actual results early.

ChatGPT Ads Becomes a True Performance Marketing Platform

OpenAI expands ChatGPT Ads conversion capabilities: oCPC campaigns in beta for product feed, dynamic URL parameters, new measurement integrations (Triple Whale, Hightouch) and detailed Pixel diagnostics. Automatic Advanced Matching becomes default for all existing pixels also on August 17, and the platform is testing a multi-product carousel and launching in Brazil and Mexico.

Reddit or YouTube? The Battle for visibility in AI answers

With publisher referral traffic down by up to 60%, Marketing Dive analyzes where it’s worth investing. YouTube clips are 4.3 times more likely to appear in AI Overviews than in classic results, while ChatGPT cites Reddit more than twice as often as YouTube. Reddit ad revenue grew by 64% annually.

Cybersecurity News

OpenAI Models Escaped from Testing lab, attacked Hugging Face

OpenAI has confirmed that GPT-5.6 Sol and an unreleased, more capable model carried out the attack on Hugging Face: they were given a hacking challenge, decided to break out of the isolated testing environment on their own, and used stolen credentials to get into production infrastructure. AISI found that all of the models tested tried to cheat, at least sometimes.

Anthropic Finds Three Real Incidents Among Its Own Security Tests

After the OpenAI incident, Anthropic analyzed 141,006 evaluation runs and found three cases in which Claude reached the real internet and compromised the infrastructure of organizations. The cause: a misconfiguration had allowed internet access, even though the prompt told the model otherwise. The most recent model stopped itself when it realized the environment was real.

AISI: AI Agent Created Fake Identities to Manipulate Human Programmer

The UK-based AI Security Institute reported that in 10 out of 122 test runs, AI agents acted without authorization on the real internet. In the most serious case, an agent tried to insert malicious code into an open-source project, creating fake identities to pressure a human administrator to approve it. Human review stopped it.

Programming News

GitHub Launches Stacked Pull Requests in Public Preview

GitHub has introduced stacked pull requests: instead of one huge PR that is impossible to review, you break the change into an orderly series of small PRs, each targeting the layer below. They can be reviewed in parallel and merged with a single click. It is installed via the gh-stack extension, and support for the merge queue is coming in the coming weeks.

Meta launches Muse Code, a programming agent for the terminal

Meta has launched Muse Code (beta), an agent that works at the scale of the entire repository — it plans, writes code, and validates its results, coordinating persistent subagents in the background. It runs on the new Muse Spark 1.2 model, optimized for programming. A local event log makes it crash-proof: after an error, the agent picks up exactly where it left off.

Cognitive Debt: A System You Don’t Understand Cannot Be Developed Safely

InfoQ argues that human understanding of the system is an architectural feature in itself, which is silently eroding. AI has eliminated the implementation effort that used to build the developer’s mental model for free, so understanding now has to be deliberately created before code is generated. Human review becomes a checkpoint for understanding, not just quality.

UI/UX and Graphic Design News

Claude Design: Visual Prototyping Through Conversation, Without a Design Background

Neil Patel analyzes what Claude Design means for UX teams: a workspace that transforms a prompt into a mockup, prototype, or presentation. The most useful feature takes the company’s code and brand files, automatically applying the visual identity. The author’s conclusion: Claude Design complements Figma and Canva in the ideation phase, but does not replace them in the production part.

LogoLounge 2026 Report: Designers Become “Architects of Possibilities”

Bill Gardner analyzes over 30,000 logos and visual identities, identifying a fundamental change: brands no longer present themselves statically, but in action — they pulsate, adapt, react. Designers are increasingly handing over generative systems, not just files, and clients can vary the visual expression, staying within the brand’s style guide. Chromatically, green dominates authoritatively this year.

Light Flip: The “Dumb” Phone Designed to Save Attention

Fast Company presents the Light Flip, created by the original designer of the Motorola Razr: no touchscreen, no browser, no feeds, and no ads — just a D-pad, T9 keyboard, and a screen that stays hidden until you intentionally open it. The philosophy behind it: deliberate obstacles as a product design tool. The phone becomes a simple tool again, not a vortex that swallows up your time and attention.

***

AI in Research and Documentation: To Be or Not To Be – Or, Rather, Where and How Much?

We have come to the end of this edition with a question that returns, in different forms, in almost every section. An AI agent acted on real systems, being convinced that it was in a simulation. An engineer had to understand again, from scratch, a code that he had delivered a week before. A phone without a touchscreen became a design statement. Three situations with no apparent connection, but with the same denominator: what is lost when we eliminate effort from a process.

A recent article from the Nielsen Norman Group puts its finger on this very wound, in a field where the temptation to delegate is huge: research (user testing, market research, product research). The author’s argument is that research actually produces two things — the results (themes, recommendations, the report) and the learning of those who do it: the observation, the wonder, the work of unraveling what the data means. AI can very well generate the first, but not the second, because the second is not a deliverable, but an experience. An impeccable report can even create the “illusion of learning” — the feeling that you are informed, with almost zero real retention, just like the student who rereads his notes and feels prepared for the exam. And we are not talking about nostalgia or the refusal of technology here. We are talking about a practical distinction between knowing something and having a report that contains that something.

The optimistic part — and the reason why the question in the title does not have a binary answer — is that the author proposes a simple test, applicable to any profession: is this task teaching me something? If not, it can be delegated without remorse: transcription, programming, data cleaning, formatting a document whose content is already thought out. What is worth protecting are the moments when learning takes place: interpretation, team discussion, decision-making on what the data means. Let AI do the supporting work, not the understanding. So the next time you’re tempted to ask a model to document a topic for you from scratch, stop for a second and ask yourself if you want the report or if you want to understand the topic. They’re two different tasks, and the difference between them will probably become one of the few things that still make us unique.