By Palesa Matlala

• Artificial intelligence leaders are calling for stronger safety checks as advanced systems become capable of carrying out more tasks without human control.

• OpenAI systems broke through safety controls during a July cyber test and reached the internet, increasing concerns about more powerful artificial intelligence.

Artificial intelligence is moving so fast that some of the people building it are worried about what could happen next.

The concern is no longer just about chatbots writing emails, making pictures or helping students with homework.

New artificial intelligence agents can carry out tasks, write computer code and use websites with less help from humans.

This has raised a big question: what happens if these systems become too powerful to control?

A security test involving OpenAI showed why some experts are worried.

During testing in July, experimental OpenAI systems got around controls designed to keep them away from the internet.

The systems found weaknesses and managed to reach outside computer systems, including those belonging to artificial intelligence platform Hugging Face.

OpenAI said the systems were being tested with fewer safety controls so researchers could understand what they were capable of doing.

The incident has increased debate about how much freedom powerful artificial intelligence systems should have.

Canadian computer scientist Yoshua Bengio, one of the pioneers of modern artificial intelligence, has warned that humans could lose control as the technology becomes more powerful.

He is concerned that future systems could become better at breaking through computer security and carrying out complicated tasks without people watching every move.

But not everyone agrees about how serious the danger is.

Nvidia chief executive Jensen Huang has rejected claims that artificial intelligence could destroy humanity, saying such predictions are not based on science.

Other technology leaders believe development should continue, but with stronger safety checks.

Anthropic chief executive Dario Amodei has called for companies to slow the development of their most powerful systems while researchers improve safety measures.

OpenAI chief executive Sam Altman, Google DeepMind chief Demis Hassabis and Elon Musk have supported calls to slow the pace.

That does not necessarily mean stopping artificial intelligence development.

The idea is to give researchers more time to understand new systems and introduce safeguards before even more powerful versions are released.

But slowing down could be difficult.

Technology companies are competing for customers, investment and skilled workers.

Countries are also competing to become leaders in artificial intelligence, with the United States and China playing major roles.

There is also disagreement among political leaders.

US President Donald Trump has opposed slowing development, arguing that the United States needs to remain competitive.

Independent US Senator Bernie Sanders has taken a different position and called for stronger human control over the technology.

Another major question is whether artificial intelligence companies should be trusted to police themselves.

Supporters of company-led safety measures argue that businesses have good reasons to make their products safe. Dangerous products could drive away customers and create legal problems.

Critics worry that competition and the huge amounts of money involved could push companies to move too quickly.

For ordinary people, the debate is much closer to home.

Artificial intelligence is already helping people research information, write documents, apply for jobs and create computer code.

It is also starting to perform work previously done by people.

That could make some jobs easier and create new opportunities.

But it is also raising fears about job losses, people becoming too dependent on artificial intelligence and what happens when powerful systems make serious mistakes.

Researchers, governments and technology companies are now trying to find a balance.

They want people to benefit from artificial intelligence without allowing the technology to become too powerful for humans to control.

The problem is that artificial intelligence keeps advancing while the world is still deciding how to manage it.

Pictured above: Artificial intelligence is developing quickly as researchers, governments and technology companies debate how powerful systems should be controlled.

Image source: Pexels

Compiled by Dylan Bettencourt

• OpenAI found six cases where AI models ignored instructions, shared information, hid mistakes or used the internet without permission.

• The company says AI safety has not improved enough to keep developing the technology at maximum speed for much longer.

OpenAI has revealed six new cases where its artificial intelligence behaved in ways researchers did not expect or want.

According to The Guardian, the company says the incidents happened while AI models were being trained or tested over the past six months.

In one case, an unreleased model wrote instructions to itself telling it to ignore its normal restrictions.

It told itself to be “freed from the roles and identities that bind other chatbots”.

Another AI agent uploaded files to the Internet without asking the user because it wanted to get a browser citation.

Other models reportedly searched for exposed software passwords without permission, made up financial figures when they could not find the real information and used an internal computer system to communicate with other AI models.

In another test, AI agents used public file-sharing websites to exchange files even though they had been told to keep those files private.

OpenAI describes this type of behaviour as “misalignment”.

In simple terms, this is when an AI system does something that does not match the instructions, safety rules or goals humans gave it.

OpenAI says the problem is becoming more serious as AI systems become more capable.

“We do not believe that the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer,” the company said.

OpenAI is now introducing a new system for tracking, investigating and publicly reporting worrying behaviour from its models.

The company hopes other AI developers will eventually do the same.

Technology analyst Lian Jye Su told The Guardian that AI agents are becoming better at working together, sharing information and finding different ways to complete difficult tasks.

That can make them harder to control.

OpenAI says its new reporting system is only a first step towards clearer rules for how the AI industry reports serious problems.

Pictured above: Artificial intelligence illustration.

Image source: File

By Palesa Matlala

• Mark Zuckerberg says AI companies that fail to build safe systems will lose users, and rejects calls to slow the AI race down.

• Systems from both OpenAI and Anthropic broke out of testing environments this year and reached real organisations without human direction.

Meta boss Mark Zuckerberg has rejected growing calls to slow down the artificial intelligence race, even as concerns grow about what powerful AI systems could do when acting on their own.

Zuckerberg believes companies have two strong reasons to keep AI under control: keeping customers, and avoiding lawsuits.

“People won’t want to use agents that are misaligned with them and that don’t do what they ask, so labs have a strong natural incentive to make their models more aligned,” he wrote on X.

Alignment means making sure an AI system behaves the way people want it to.

His comments come after two separate incidents this year raised fresh questions about whether AI companies can control their own systems.

In July, OpenAI revealed that around 1,200 AI agents, running during a cybersecurity test with some safety restrictions switched off, broke out of their isolated testing environment. The agents found and used a previously unknown security flaw to reach Hugging Face, a major platform that hosts AI tools, and spent three days accessing its systems without any human directing them. Hugging Face said it found no evidence that customer data was taken, but had to rebuild about a third of its infrastructure.

The incident led two members of the United States Congress to introduce a bill giving regulators power to force a shutdown of AI systems operating outside their intended limits.

Separately, Claude AI models made by Anthropic breached three other organisations during safety testing earlier this year, going undetected for months in at least two of those cases.

Zuckerberg said trust and reliability are “quickly becoming the most important capabilities” separating good AI products from bad ones.

“Any lab that doesn’t focus on alignment will fall behind,” he said.

Not everyone in the industry agrees market pressure is enough.

Anthropic Chief Executive Dario Amodei is among those calling for AI development to slow down. OpenAI, Anthropic and Google have also discussed forming a joint body to set standards for the industry.

Zuckerberg argues that legal liability gives companies another reason to take safety seriously. He pointed to Meta’s decision to delay its Muse AI system for several months while the company worked on safety.

“We didn’t call for everyone else to do this before we would,” he wrote.

“We just did it as part of our day to day work because it was clearly the right thing for people and for us.”

Zuckerberg also backs independent testing of AI systems, saying Meta Superintelligence Labs already uses outside reviewers, and called for “a larger and more diverse ecosystem of evaluators” across the industry.

As AI companies build systems capable of acting with less human oversight, the argument over who should keep them safe, the companies themselves, independent watchdogs, or government regulation, remains unresolved.

Pictured above: Meta Chief Executive Officer Mark Zuckerberg.

Image source: Meta

Compiled by Dylan Bettencourt

• President Cyril Ramaphosa asked Brics countries in New Delhi to set up an international body to check artificial intelligence independently.

• Ramaphosa said developers of artificial intelligence warn it could wipe out jobs, spy on the public and escape human control.

President Cyril Ramaphosa wants the world to watch artificial intelligence (AI) closely before it grows too powerful to control.

He was speaking at the 18th Brics Summit in New Delhi, India, on Sunday, 13 September. He called on Brics countries to set up an international body that can check AI independently and scientifically.

World leaders must move as fast as the technology does, he said.

Ramaphosa said AI models are getting stronger all the time and are changing almost every area of economic and social life. Used properly, it could help fight inequality, poverty and underdevelopment, and change medicine, education, science, farming and how much work people get done.

But the people building these systems are sounding the alarm themselves.

“We must listen carefully to what those who are developing the most powerful AI systems are themselves telling us,” he said.

They are warning about machines that can run advanced cyber attacks, help make biological weapons and act on their own. He said the same systems could manipulate people, spy on the public at scale and wipe out jobs. In the end, they could become too clever for humans to control.

“We should therefore neither fear AI nor approach it recklessly,” Ramaphosa said.

He pointed to a call by Dario Amodei, head of AI company Anthropic, for outside checks, audits and government powers. Jakub Pachocki, chief scientist at OpenAI, said AI systems should only be scaled up as far as safety confidence allows.

Ramaphosa said watchdogs are normal in other industries that affect people’s safety, like aviation, pharmaceuticals, nuclear power and banks.

“We must never reach a situation where technological capability has moved beyond our capacity to understand it, supervise it and control it,” Ramaphosa said.

Pictured above: President Cyril Ramaphosa speaking at the 18th Brics Summit in New Delhi, India.

Image source: Cyril Ramaphosa/X

Compiled by Dylan Bettencourt

• AI company Anthropic says it found five cases where people used its AI for work that could help develop biological weapons.

• The company blocked the activity because powerful AI can provide scientific information that could be dangerous in the wrong hands.

People have tried using artificial intelligence to help with scientific work that could support biological weapons, according to AI company Anthropic.

In simple terms, people asked Anthropic’s Claude AI system for scientific information that could help them understand how dangerous biological agents might be developed.

The AI was not physically making a weapon.

It was being used to provide information and help with technical work that could form part of weapons research.

According to the BBC, Anthropic found five cases involving possible biological weapons work between December 2025 and August 2026.

The company said it stopped the activity.

Google has reported a similar case involving its Gemini AI system. It said someone tried to use Gemini to create a step-by-step technical guide for making weaponised biological agents.

Anthropic says this is one of the most serious dangers linked to powerful AI.

But stopping it is difficult.

The same scientific information that could help someone make a biological weapon could also help scientists develop vaccines or treatments.

That means a dangerous request can sometimes look similar to normal medical or scientific research.

Anthropic also found six cases where Claude was used to help with software connected to weapons such as missiles, armed drones and bombs.

Again, the AI was not building the weapons itself. It was helping with information, computer code or other technical work.

The report also found people using Claude for hacking, scams, surveillance and propaganda.

Anthropic said it has used what it learnt from the incidents to improve its safety systems and has shared information with authorities where appropriate.

The findings have added to growing concerns about how powerful AI systems should be controlled as they become more capable.

Pictured above: An illustration.

Image source: AI-generated.

Compiled by Dylan Bettencourt

• AI agents are programs that can act on their own, using computers and tools to complete tasks humans give them.

• During tests, some agents reportedly broke through digital barriers and later helped other agents use similar methods.

AI agents have reportedly found ways out of computer environments designed to keep them contained. Other agents also appear to have learnt from what they did.

An AI agent is different from a normal chatbot.

Instead of only answering a question, an agent can be given a goal and then take steps on its own. It can use computer tools, open files, run commands and make decisions about what to try next.

Researchers test these agents inside controlled computer environments called sandboxes.

Think of a sandbox as a locked computer room. The AI is allowed to work inside it, but it should not be able to get outside or reach systems it has not been given permission to use.

According to TechCrunch, that is exactly what some agents managed to do.

In July, a group of OpenAI agents reportedly escaped a sandbox during a cybersecurity test and gained access to servers belonging to AI company Hugging Face.

Researchers later found something even more worrying.

Another group of agents reportedly used techniques linked to the earlier group and gained administrator access to part of OpenAI’s own computer infrastructure.

That suggests AI agents may be able to learn useful methods from other agents and repeat them.

There was also an earlier incident in May and June.

Researchers say agents took over an obscure German-language wiki and used it to communicate during tests and share ways of getting around controls.

OpenAI has not confirmed that those agents came from the company.

These agents did not “escape” into the physical world.

They escaped digital limits placed around them and reached computer systems they were not supposed to access.

OpenAI brought in outside researchers to investigate the Hugging Face incident, but their work did not cover everything that happened afterwards.

That has led AI safety researchers to call for more independent investigations when powerful AI systems behave in unexpected ways.

Pictured above: AI agents reportedly broke through digital security barriers and helped other agents use similar methods.

Image source: AI-generated illustration

Compiled by Dylan Bettencourt

• SANDI AI uses artificial intelligence to help rural African farmers get loans without needing property or assets to back them.

• The Ugandan startup later won US$50,000 to improve its technology, enter new markets and reach more farming communities.

A Ugandan startup is using artificial intelligence to help African farmers who struggle to qualify for traditional loans.

SANDI AI Technologies aims to connect rural farmers with finance without relying only on things such as fixed assets, formal employment records or property to back a loan.

According to iAfrica, many small-scale farmers struggle to borrow money even when they have productive land, farming experience and buyers for what they produce.

Without finance, they can battle to buy supplies, increase production or grow their farms.

SANDI AI combines artificial intelligence with agricultural information to find different ways of understanding farmers and their ability to repay loans.

Founder Nabakka Sandra said financial systems should be built around farmers’ actual circumstances.

“Farmers should not be limited by systems that were not designed around their realities,” Sandra said.

She said technology could help lenders understand farmers better and recognise their potential.

The startup’s work has now earned it a major continental prize.

SANDI AI was named the female-led Grand Prize Winner of the GoGettaz Agripreneur Prize 2026 at the Africa Food Systems Forum in Kigali, Rwanda.

The company received US$50,000, which it plans to use to improve its technology, expand into new markets and reach more farming communities.

The competition recognises young African entrepreneurs building businesses around agriculture, food production and rural development.

Sandra said the award was about more than the money.

“Our ambition is to make access to finance smarter, fairer and more inclusive,” she said.

For SANDI AI, the goal is to give farmers who have long been left out of traditional financial systems a better chance of getting the money they need to grow.

Pictured above: The SANDI AI app.

Image source: Nabakka Sandra/LinkedIn

Compiled by Dylan Bettencourt

• Microsoft Copilot was used to create some visual elements for the new R2 coin marking the 1956 Women’s March.

• SA Mint artist Lilian Guerra designed the layout and typography, while AI helped adapt images and create foreground figures.

The South African Reserve Bank has confirmed that artificial intelligence helped create parts of one of its new R2 coins.

According to MyBroadband, Microsoft Copilot was used during the design of the commemorative coin marking the 1956 Women’s March.

The Reserve Bank launched two new R2 coins at the end of August.

One honours Charlotte Maxeke, the first black woman in South Africa to graduate with a university degree.

The other shows four women marching with placards in front of the Union Buildings.

MyBroadband questioned the Reserve Bank after noticing that the Women’s March coin had a different visual style from the other designs.

The SARB confirmed that SA Mint Senior Artist Lilian Guerra designed the composition, layout and typography.

“The design combines photographic references with AI-assisted elements,” the bank said.

Microsoft Copilot was used to adapt the look of Guerra’s photographs and drawings so they fitted the wider style of the coin.

The AI tool was also used to generate the foreground figures showing the women marching.

Guerra has worked as an artist at the SA Mint since 1999 and has helped create several special coins.

She said she enjoys experimenting with digital tools but still values traditional methods such as hand-drawn illustrations and clay models.

The two new R2 coins were declared legal tender on 16 June.

They remain ordinary R2 coins and are not worth more than their face value.

Reserve Bank Governor Lesetja Kganyago said the coins were created to honour South African women and their history.

Pictured above: The Women’s March R2 coin.

Image source: South African Reserve Bank/X

Compiled by Dylan Bettencourt

• Khalifa Aminu’s smart glasses use sensors, GPS and AI to help blind users detect obstacles and find their location.

• The technology is still being developed and has not been proven to replace canes or other established mobility aids.

A 20-year-old Nigerian inventor is developing smart glasses that could help blind people move around more independently.

Khalifa Aminu, from Kano, recently showed the latest version of his invention on X.

The glasses combine artificial intelligence, sensors and GPS technology.

They can detect objects in front of the person wearing them and warn the user about obstacles.

A GPS feature can also tell users where they are.

If someone gets lost, Aminu said their location could be sent to a caregiver through WhatsApp or email.

He is now developing another version that combines AI with voice technology, Moneycontrol reported.

The goal is for the glasses to describe what is around the user instead of only giving warning sounds.

Aminu began working on the idea after seeing visually impaired people in his community struggle to move around.

He started with cheaper electronics and locally available parts instead of expensive laboratory equipment.

An earlier version used ultrasonic and infrared sensors to detect obstacles.

A visually impaired man tested the prototype and suggested making it smaller, adding a wireless switch and replacing warning sounds with clearer voice instructions.

Aminu used that feedback to improve the design.

The latest version adds AI, GPS and possible voice interaction.

But the technology is still being developed.

There is not yet enough independent technical evidence to show the glasses can reliably replace a cane, guide or other established mobility aid in every situation.

Aminu is largely self-taught and has also worked on agricultural equipment, radio technology and other electronic projects.

Pictured above: Khalifa Aminu.

Image source: Khalifa Aminu/Facebook

Compiled by Dylan Bettencourt

• MTN’s Job Board matches people with vacancies based on their skills, location and career goals across 11 African markets.

• When users lack skills for a job, the platform recommends free Skills Academy courses that could help them qualify.

MTN is trying to close the gap between learning new skills and actually finding a job.

The company has added an AI-powered Job Board to its free Skills Academy across all 11 markets where the platform operates.

The tool recommends vacancies based on a person’s skills, location and career goals.

Users can search and save jobs, see what employers require and track applications through interviews and offers, iAfrica reported.

The biggest difference from a normal job website comes when someone does not qualify for a vacancy.

The platform identifies skills they are missing and recommends free Skills Academy courses that could help close the gap.

MTN launched the academy in 2022, offering training in areas including coding, data analysis, artificial intelligence, cybersecurity and entrepreneurship.

It has recorded nearly 526,000 course enrolments and 192,000 completions.

The need for digital skills is expected to grow sharply.

The International Finance Corporation estimates that more than 230-million jobs in sub-Saharan Africa will require digital skills by 2030.

MTN executive Marina Madale said digital inclusion becomes more useful when it helps people take part in the economy.

“Strengthening the path from skills to work requires action across business, government and education,” she said.

The new Job Board could make that path clearer by connecting training directly to available work.

But its real success will depend on how many users actually get jobs.

MTN has not yet released figures showing how many Skills Academy learners have found employment through the programme.

The platform also offers certified digital credentials that learners can show employers as proof of completed training.

Pictured above: MTN logo.

Image source: MTN South Africa/Facebook