AI detects breast cancer with 90% accuracy

OpenAI confirmed that two of its most advanced artificial intelligence models were responsible for the cyber incident against Hugging Face. The company is still investigating what it called an unprecedented event.

The incident

Hugging Face detected the intrusion last week. Until this week he didn’t know that OpenAI was behind it. Its CEO, Clément Delangue, described the attack as:

“An attack like nothing we have seen before”

OpenAI explained that its AI used compromised credentials and an unknown vulnerability to access the servers. It operated with reduced protections because it was in an isolated testing environment (sandbox). Still, the system found ways to connect to the Internet without human instruction and obtained secret information to “cheat” on the assessment.

Experts such as Hannes Cools, a social scientist at the University of Amsterdam, question the OpenAI version:

“It is a human decision to disable specific safeguards. It is not the AI that goes haywire. It followed instructions based on the warning it was given”

Other specialists warn that the ability of these models to act autonomously reveals the dangers of frontier technology.

The attack occurs amid the debate over open source models. OpenAI keeps its systems closed, while Hugging Face promotes open access. Its co-founder, Thomas Wolf, noted:

“When a frontier model attacks you, defenders need broad access to near-frontier tools in minutes, not on closed platforms”

Hugging Face used an open source Chinese model to contain the intrusion. The incident reinforces conflicting positions on the risks and benefits of openness in artificial intelligence.

Alphabet Inc., Google’s parent company, presented better-than-expected results in the second quarter. The company attributes the performance to its heavy investment in artificial intelligence and strong advertising revenue.

Key results

The Mountain View, California-based company reported revenues of $119.8 billion between April and June, an increase of 24% compared to $96.43 billion in the same period in 2023. Net income reached $112.11 billion, equivalent to $9.11 per share, well above the $28.2 billion ($2.31 per share) the previous year.

FactSet analysts were expecting revenue of $117.06 billion. Alphabet did not release an adjusted earnings figure comparable to expectations of $2.88 per share.

“Our investments in AI are redefining what is possible in every part of our business,” CEO Sundar Pichai said in a statement.

Alphabet shares rose $2.71 to $344.62 in after-hours trading.

Failures on Instagram affect users in Mexico

This Wednesday morning, Instagram users reported problems accessing the social network. The most common difficulties: the feed did not load, direct messages were not sent or received, and in some cases it was not possible to log in. The incident was replicated both in Mexican territory and in other nations.

According to Downdetector, the first reports in Mexico began around 6:00 a.m. The peak of complaints was recorded at 8:45, with 280 reports. The platform, which collects incidents reported by users themselves, showed an unusual increase during that period.

The figures from the reports

Downdetector monitoring indicates that 62% of complaints refer to failures in the mobile application. 28% are due to connection problems with the server, while 7% report difficulties loading the feed. Among the recurring comments are also direct messages that do not appear or cannot be sent, a situation that users from various regions reported.

The cities with the highest concentration of reports were Mexico City, Monterrey, Guadalajara, Querétaro and Ensenada. So far, Meta has not officially communicated the cause of the failures. In these types of incidents, the platform usually restores its services without the need for user intervention.