In Black Hat and Def Con, Amnesty International was infiltrated, a personal guard, and a time goal

Las Vegas feels in August as if it were another planet – heating heat, flashing lights, and continuous sinks of gambling machines. This other atmosphere has moved to the two conferences that I attended last week: Black hat and Def Con, one of the largest security conferences in the year and “moral” piracy, where advanced security research is presented; Snements are racing to expose defects in everything from chat tools from artificial intelligence to energy networks; Governments, companies and amateurs exchanging notes and learning the latest technology to combat threats.

I was there because artificial intelligence now occupies a strange place in the world of security – it is a weak goal, and is often threatened with harmful artificial intelligence; Armed defender uses the same technology to protect systems and networks from bad actors; A offensive player, who was published to investigate weaknesses or carry out attacks (illegally in criminal hands). The sound is confusing? It is – and this contradiction was fully displayed in the black hat of companies, as well as Def Con, often referred to as “Hacker Summer Camp”.

Below are three of my favorite fast food. Conferences:

• Chatgpt agents can be hacked. In Black Hat, researchers from the security company Zenity showed how infiltrators can take advantage of the new OpenAi connector feature, which allows Chatgpt data withdrawing data from applications such as Google Drive, SharePoint and GitHub. Concept proof attack, called AgentflayerUse a “poisoned” with an innocent appearance with hidden instructions to make Chatgpt looking at the victim’s files to obtain sensitive information and send it to the attacker. Kicker: It does not require any user clicks or downloads – this video He explains how it was done. Openai’s defect was fixed after alert, but the episode emphasizes the increasing risk as artificial intelligence systems are associated One informed him In June), Chatbots trick to make bids to the infiltrator without the user’s awareness.

• Amnesty International can protect our most important infrastructure. This idea was the driving force behind the AI Cyber (AIXCC) challenge for two years, which cost developers teams to build tools of obstetric intelligence to find weaknesses in software and fix them in the code that indicates everything from banks and hospitals to public facilities. The competition-which was conducted by DarPa in partnership with ARPA-H- at Def Con for this year, where the winners showed independent artificial intelligence systems capable of securing an open source program that supports a lot of critical infrastructure in the world. The first three teams will receive $ 4 million, 3 million dollars and 1.5 million dollars, respectively, for their performance in the finals.

• Claude Ai from Anthropor has proven that it can sometimes match the elite infiltrators but still struggle with difficult problems. Kane Lucas, a member of the Antarbur border border team, presented a wonderful case on how the company calmly entered the artificial intelligence into seven major cybersecurity security competitions – usually dominated by human infiltrators and security professionals – to know how to accumulate it. Claude often landed in the upper quarter of competitors and showed a prominent speed on simple challenges, sometimes identical to the elite human difference, but it failed to the most difficult problems. There were also unique dodges for Amnesty International – such as ASCII ART decreased or launched to philosophical wandering during long tasks. Anthropor says these experiences highlight all the offensive capabilities of Amnesty International (by reducing the skill and the cost that it objects to by launching attacks) and their defensive promise, while emphasizing the continuous limits such as long -term memory and complex thinking.

Also: within just a few weeks, I will go to Park City, Utah to participate in our annual conference technology in Montage Deir! The space is limited, so if you are interested in joining me, Sign here. I highly recommend: There is a wonderful group of speakers, including Ashley Kramer, chief revenue in Openai; John Werner, President and CEO of Walmart Us; Tony Show, Founder and CEO of Doordash; And a lot, a lot!

However, here is more news of artificial intelligence.

Sharon Goldman
[email protected]
Sharongoldman

Amnesty International in News

China requests companies to avoid NVIDIA H20 chips after Trump ends the ban, but takes revenue. according to BloombergBeijing advised calm Chinese companies, especially those participating in government or national security projects, to get rid of H20 processors in NVIDIA, which led to the efforts of the American chip maker’s efforts to restore billions in the lost sales of China after the Trump administration lifted an effective ban on such exports. The guidance, which has been delivered in recent weeks to a group of companies, does not completely prohibit it but indicates that there is no power to use the H20 in sensitive contexts. The news comes after President Trump announced that the United States government will take 15 % of the revenue that Nafidia’s chip maker withdraws from sales in China from H20 chips.

Students flow to university courses in artificial intelligence. When artificial intelligence converts the workplaces, increasing numbers of professionals and students follow certificates, certificates and advanced training in artificial intelligence to enhance skills and work prospects. the Washington Post I mentioned Universities such as the University of Texas in Austin, University of Michigan in Derborn, San Diego University, and Massachusetts Institute of Technology are witnessing enrollment in artificial intelligence programs, which are fueled by the rise of Shattab and the employer to erase the literacy of Amnesty International. Training courses range from technical training in algorithms and building models to broader education in creativity, ethics and problem solving. Corporate partnerships, such as Meta Inving Bustructors, reflects the urgency of industry, while new initiatives in schools such as Ohio aim to make AI’s literacy globally by 2029. Workers see that artificial intelligence skills are a competitive advantage-PWC reports for wages for financial employees of artificial intelligence-and equipment Experts are optimistic and they are experts to adapt alone.

Antarubor Claude will offer Amnesty International to the United States government for one dollar. Reuters I mentioned ANTPRORPROR will offer a Claude AI model to the United States government for one dollar. This announcement comes days after Openai Also displayed chatgpt To the United States government for one dollar, while Gemini from Google was also added to the list of the government’s artificial intelligence sellers. “The leadership of artificial intelligence in America requires our governmental institutions to have the most successive consecutive artificial intelligence tools,” said Dario Amani, CEO.

Eye on artificial intelligence research

Reports say that artificial intelligence training to obtain the needs of the huge power

According to axios, a New report From the Electrical Power Research Institute and APOCH AI reveals the amazing energy requirements to train large “Frontier” models: by 2028, individual training projects can require 1-2 GW of energy, and by 2030 the range may extend to 4-16 GW-with a high end representing approximately 1 % of US electricity capacity. In general, it is expected that the energy needs of the United States (which cover both training and inference) will rise from 5 GB today to 50 GB by 2030 according to the report, giving politics and high -level technology companies basic visions to help planning infrastructure and energy strategies despite the great uncertainty.

Our wealth

Exclusively: The deep raises 35 million dollars, as Sequoia supports its ambitious offer to become Salesforce of Ai Search Sharon Goldman

How to deal with Apple and Trump’s Pentagon, which has turned into MPER MERER EARTER materials into red stocks – By Jordan Blum

It explains the Openai Open Source axis – Nicholas Gordon

The endless thirst of artificial intelligence is the driving of the natural gas boom in Apalashia – and the shares of industry flourish with it – By Jordan Blum

You have a calendar

September 8-10: Fortune Brainstorm Tech, Park City, Utah. Apply to attendance here.

6-10 October: world Amnesty International The week, Amsterdam

October 21-22: Teddy San Francisco. Apply to attendance here.

2-7 December: NeuPIPs, San Diego

December 8-9: Fortune Brainstorm Ai San Francisso. Apply to attendance here.

Brain food

An apple for the teacher? These days, it is Amnesty International. A new article in Atlantic Ocean It focuses on the other side of the use of artificial intelligence in schools: Chatbots such as ChatGPT may be popular with high school students, but teachers also embrace artificial intelligence to save time in gradient, planning lessons, and papers, with platforms such as Magicschool AI that are now used by millions of American teachers. Educational areas-some are prohibited from artificial intelligence, others merge tools such as Google Gemini in the classroom-but examples of Houston show the risk of weak quality materials generated by artificial intelligence. The federal policy of adopting artificial intelligence is pushed widely, with great financing and microsoft partnerships, Openai and anthropology. However, consensus lacks how to balance the efficiency of artificial intelligence while maintaining students’ critical thinking skills, leaving schools at a crossroads where increasing dependence on artificial intelligence can now consolidate them deeply in future education.

This is the online version of Eye on AI, Fortune’s weekly newsletter on how to form AI for the future of business. Subscribe for free.