OpenAI is reportedly working on a new project named “Strawberry,” designed to enhance the reasoning capabilities of its AI models. Based on the mysterious QStar advance, Strawberry aims to enable AI to perform long-horizon tasks (LHT), requiring it to plan and execute actions over extended periods. This initiative could significantly advance AI's ability to autonomously navigate the internet and conduct deep research. Although the specifics of Strawberry remain confidential within OpenAI, its goal is to elevate the intelligence of AI models closer to or beyond human-level capabilities by employing post-training techniques similar to the "Self-Taught Reasoner" method from Stanford.

New Rumor of an OpenAI Next Level Reasoning Breakthrough NextBigFuture

Google is nearing a $23 billion acquisition deal for Wiz, a cybersecurity startup. This move signifies Google's intention to expand its cybersecurity capabilities and highlights the growing emphasis on protecting digital assets amid increasing threats. The acquisition aligns with the tech giant's strategic focus on enhancing security measures for its cloud services and enterprise clients.

Google Near $23 Billion Deal for Cybersecurity Startup Wiz Wall Street Journal

AT&T reported a cyberattack on its Snowflake environment that compromised data of nearly all of its wireless customers, affecting approximately 110 million individuals. The breach included records of customers' calls and text messages over a six-month period ending October 31, 2022, and January 2, 2023, but did not expose the content of the communications or personal identifiers. However, it did include phone numbers, interaction counts, and call durations. The incident was part of a broader wave of attacks on Snowflake customers, exploiting stolen credentials from infostealer malware. AT&T detected the breach on April 19 and acted swiftly, involving cybersecurity experts and cooperating with law enforcement. The company has enhanced its cybersecurity measures and will notify affected customers, while the FBI and DOJ granted delays in disclosure due to national security concerns.

Massive Snowflake-linked attack exposes data on nearly 110M AT&T customers Cybersecurity Dive

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) released a report highlighting significant shortcomings in cybersecurity practices within a U.S. civilian executive branch agency. The report follows a 2023 red-teaming exercise where CISA employed tactics resembling those of nation-state threat actors and successfully compromised the agency's network, gaining access through a known Solaris vulnerability and phished Windows credentials. The red team remained undetected for a significant period, even managing to eavesdrop on the blue team's communications. Key recommendations from CISA include streamlining incident response, avoiding dependence on known indicators of compromise, and enhancing log monitoring and analysis for better attack comprehension and defense.

CISA sees red over government cybersecurity exercise scmagazine