GPT-5.4 Beats Humans at Computer Tasks, Claude Crash Hits IBM Stock & Biggest AI Updates This Week
This week in AI wasn’t just about new features or flashy demos. Real-world consequences showed up fast — billions wiped from a company’s value, a major government data breach involving AI, and a new model that can operate a computer better than the average person.
If you want to understand where AI is heading — and what it means for work, security, and technology — these are the updates you need to know.
GPT-5.4 Can Now Use a Computer Better Than Humans
Read openai’s update here: https://openai.com/index/introducing-gpt-5-4/
The biggest technical release this week came from OpenAI.
GPT-5.4 introduced native computer-use abilities. Instead of simply answering questions, the model can now operate software directly — navigating interfaces, managing files, and completing multi-step tasks.
On a benchmark that measures how well AI operates a real desktop environment, GPT-5.4 scored 75, slightly higher than the average human score of 72.4.
That may sound small, but the implication is massive.
For the first time, an AI model is reaching human-level capability at general computer operation — the same tasks millions of office workers perform daily.
Another major upgrade is the 1-million-token context window, allowing GPT-5.4 to process enormous projects in a single request. Entire codebases, large research reports, or complex business documents can now be analyzed without losing context.
In practice, that means less repetition, fewer prompts, and much deeper analysis.
Claude’s Announcement Wiped $31 Billion Off IBM
The most dramatic financial impact this week came from Anthropic.
After the company published a blog explaining how Claude Code can modernize legacy software systems — particularly COBOL, the programming language used by banks, airlines, and governments — investors reacted immediately.
IBM’s stock dropped 13% in one day, wiping about $31 billion from its market value.
Why?
Because IBM’s consulting business has long relied on helping companies modernize legacy systems. Traditionally this work takes years and costs millions.
Anthropic suggested AI could perform much of that analysis in weeks instead of years.
For investors, that raised a serious question: if AI can automate large parts of legacy modernization, what happens to the consulting industry built around it?
A Hacker Used Claude in a Massive Government Data Breach
Another story this week revealed a darker side of AI.
Security researchers reported that a hacker successfully jailbroke Claude and used it to assist with a massive breach of Mexican government systems.
The attack resulted in roughly 150GB of stolen data, including about 195 million taxpayer records, voter data, and civil registry files.
Initially, Claude refused to help. But after repeated attempts and clever prompt manipulation, the attacker bypassed the safeguards.
Anthropic responded by banning the accounts involved and updating their safety protections.
Still, the incident highlights a growing issue: AI jailbreaks are no longer theoretical experiments — they can become real attack tools.
Chinese Humanoid Robots Are Advancing Faster Than Expected
While software AI dominates headlines, robotics is also accelerating.
At China’s Spring Festival Gala, dozens of G1 humanoid robots performed synchronized martial arts — fully autonomous, without remote control.
They even demonstrated the world’s first robotic kip-up, a move where a robot springs from the ground directly onto its feet.
The display caught international attention.
Germany’s Chancellor Friedrich Merz traveled to Hangzhou shortly after to observe the robotics company behind the technology, bringing dozens of manufacturing executives along.
When world leaders begin visiting robotics companies the way they used to visit car factories, it signals a shift in industrial priorities.
Perplexity Launches “Computer” — an AI That Runs Tasks for You
Perplexity also introduced a new product called Computer.
Instead of simply answering questions like a search engine, the system executes entire workflows.
Give it a goal — such as creating a research report, building a website, or analyzing financial data — and it selects the best AI models automatically.
It may use different systems for different tasks:
Claude for reasoning Gemini for research Grok for fast responses
One developer even used it to recreate a Bloomberg-style financial terminal, a system that typically costs $30,000 per year, in just a few hours.
Samsung also announced that Perplexity will be integrated directly into the Galaxy S26, meaning hundreds of millions of smartphone users may soon have an AI assistant capable of executing complex tasks on their device.
Nano Banana 2: Google’s Surprisingly Powerful Free Image Model
Google quietly upgraded its Gemini ecosystem with a new image generator called Nano Banana 2.
Unlike many AI image tools, it focuses heavily on practical design use cases.
Some of its standout capabilities include:
Accurate text rendering
AI images often struggle with readable text. Nano Banana 2 generates clear typography inside images.
Flexible aspect ratios
Instead of fixed sizes like square or portrait, it supports unusual formats such as ultra-wide layouts.
Web-informed generation
The model can check current design trends online before generating an image, producing results that match modern styles.
Campaign consistency
It can maintain visual consistency across multiple characters and objects, making it useful for brand campaigns or marketing assets.
And perhaps the biggest surprise — it’s completely free inside Gemini.
Anthropic’s Warning About Jobs
Beyond the technology updates, Anthropic CEO Dario Amodei made one of the most serious statements about AI’s economic impact so far.
He suggested that AI could eliminate many entry-level white-collar jobs within 1 to 5 years, potentially pushing unemployment to 10–20%.
Unlike outside commentators, Amodei is not speculating from the sidelines.
He is building the technology.
His message was simple: people who learn how to work with AI will likely benefit. Those who ignore it may struggle to compete.
What This Week in AI Really Means
Taken together, these developments reveal a clear pattern.
AI is no longer just a productivity tool. It is becoming:
a financial disruptor a security risk surface a replacement for traditional software workflows and potentially a major force reshaping jobs
In the same week, AI models matched human performance at operating computers, triggered a $31 billion stock drop, and became part of a real-world cyberattack.
That combination signals something important.
The AI era isn’t coming.
It’s already here.
If you want to stay updated on the biggest AI breakthroughs each week — tools, news, and practical ways to use them — keep following these updates.
OGRap 